SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2003.12039 · 2020

RAFT: Recurrent All-Pairs Field Transforms for Optical Flow

Jia Deng, Zachary Teed

arXiv · PDF · Open in the Atlas

Code that ran

41 of 66 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (15 of them constructed an object rather than computing a result; 3 honoured the contract we drafted, 0 violated it and 22 have no contract verdict; 16 failures of our instrument; 25 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2003.12039.json · badge: /api/ran/2003.12039.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
osmr/pytorchcv community
Papers with Code archive
17 ran · 13 no recorded run
plusgood-steven/id-blau community
Papers with Code archive
13 ran · 3 no recorded run
esakak/sevc community
Papers with Code archive
4 ran · 4 no recorded run
pytorch/vision community
Papers with Code archive
3 ran · 0 no recorded run
wpr1018001/rethinking-low-quality-optical-flow community (classifier label)
classifier basis not recorded on the link
2 ran · 0 no recorded run
aliciachenw/PIPsUS community
Papers with Code archive
1 ran · 1 no recorded run
duke-vision/optical-flow-active-learning-release community
Papers with Code archive
1 ran · 0 no recorded run
princeton-vl/RAFT marked official in Papers with Code
Papers with Code archive
0 ran · 3 no recorded run
copy not recorded community
not in the paper's declared repositories
0 ran · 1 no recorded run
FunctionStatusWhere it lives
BasicEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("89caa833c87120d1")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

BasicMotionEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("269ca6f91ab3d7bf")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

BasicMotionEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
aliciachenw/PIPsUS/nets/update.py
code served (permissive licence) · get_code("073e918042b41586")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, BasicMotionEncoder: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

class BasicMotionEncoder(nn.Module):

    def __init__(self, corr_levels=4, corr_radius=3):
        super(BasicMotionEncoder, self).__init__()
        cor_planes = corr_levels * (2 * corr_radius + 1) ** 2
        self.convc1 = nn.Conv2d(cor_planes, 256, 1, padding=0)
        self.convc2 = nn.Conv2d(256, 192, 3, padding=1)
        self.convf1 = nn.Conv2d(2, 128, 7, padding=3)
        self.convf2 = nn.Conv2d(128, 64, 3, padding=1)
        self.conv = nn.Conv2d(64 + 192, 128 - 2, 3, padding=1)

    def forward(self, flow, corr):
        import torch.nn.functional as F
        cor = F.relu(self.convc1(corr))
        cor = F.relu(self.convc2(cor))
        flo = F.relu(self.convf1(flow))
        flo = F.relu(self.convf2(flo))
        cor_flo = torch.cat([cor, flo], dim=1)
        out = F.relu(self.conv(cor_flo))
        return torch.cat([out, flow], dim=1)

encoder = BasicMotionEncoder(corr_levels=4, corr_radius=3)

batch_size = 2
height = 4
width = 3

flow = torch.randn(batch_size, 2, height, width)

corr_levels = 4
corr_radius = 3
cor_planes = corr_levels * (2 * corr_radius + 1) ** 2
corr = torch.randn(batch_size, cor_planes, height, width)

ARGS = {
    'corr_levels': 4,
    'corr_radius': 3,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type BasicMotionEncoder — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a BasicMotionEncoder output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
BasicUpdateBlock Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("405f48802a21a4f4")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

BottleneckBlock Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("3ee0a69d26057e27")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

ConvBlock Ran
ran on a synthesized input · constructed an object; no result was computed
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("c69c35f159717fe0")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, ConvBlock: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

def lambda_batchnorm2d(eps: float=1e-05):
    """
    Create lambda-function generator for nn.BatchNorm2d normalization layer.
    """
    return lambda num_features: nn.BatchNorm2d(num_features=num_features, eps=eps)

def lambda_relu(inplace: bool=True):
    """
    Create lambda-function generator for nn.ReLU activation layer.
    """
    return lambda: nn.ReLU(inplace=inplace)

class Swish(nn.Module):
    """
    Swish activation function from 'Searching for Activation Functions,' https://arxiv.org/abs/1710.05941.
    """
    def forward(self, x):
        return x * torch.sigmoid(x)

class HSwish(nn.Module):
    """
    H-Swish activation function from 'Searching for MobileNetV3,' https://arxiv.org/abs/1905.02244.
    """
    def __init__(self, inplace: bool=False):
        super(HSwish, self).__init__()
        self.inplace = inplace

    def forward(self, x):
        return x * torch.nn.functional.relu6(x + 3.0, inplace=self.inplace) / 6.0

class HSigmoid(nn.Module):
    """
    Approximated sigmoid function from 'Searching for MobileNetV3,'
    https://arxiv.org/abs/1905.02244.
    """
    def forward(self, x):
        return torch.nn.functional.relu6(x + 3.0, inplace=True) / 6.0

ARGS = {
    'in_channels': 3,
    'out_channels': 64,
    'kernel_size': 3,
    'stride': 1,
    'padding': 1,
    'dilation': 1,
    'groups': 1,
    'bias': False,
    'normalization': lambda_batchnorm2d(),
    'activation': lambda_relu()
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ConvBlock — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a ConvBlock output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
CorrBlock Ran
ran on a synthesized input
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("fcdbfcdd43f23a80")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

CorrCalculator Ran
ran on a synthesized input · constructed an object; no result was computed
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("304e803cf619d43e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, CorrCalculator: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Create small feature maps for testing
# fmap1 and fmap2 need to have shape (batch, channels, height, width)
batch_size = 2
channels = 4
height = 8
width = 8

fmap1 = torch.randn(batch_size, channels, height, width)
fmap2 = torch.randn(batch_size, channels, height, width)

radius = 2
num_levels = 4

ARGS = {
    "fmap1": fmap1,
    "fmap2": fmap2,
    "radius": radius,
    "num_levels": num_levels,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type CorrCalculator — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a CorrCalculator output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
Correlation Ran
ran on a synthesized input · constructed an object; no result was computed
duke-vision/optical-flow-active-learning-release/models/correlation_native.py
code served (permissive licence) · get_code("1603237c32f607ff")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, Correlation: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

torch.manual_seed(0)

class Correlation(nn.Module):

    def __init__(self, max_displacement=4, *args, **kwargs):
        super(Correlation, self).__init__()
        self.max_displacement = max_displacement
        self.output_dim = 2 * self.max_displacement + 1
        self.pad_size = self.max_displacement

    def forward(self, x1, x2):
        B, C, H, W = x1.size()
        x2 = F.pad(x2, [self.pad_size] * 4)
        cv = []
        for i in range(self.output_dim):
            for j in range(self.output_dim):
                cost = x1 * x2[:, :, i:i + H, j:j + W]
                cost = torch.mean(cost, 1, keepdim=True)
                cv.append(cost)
        return torch.cat(cv, 1)

# Create valid arguments for Correlation
x1 = torch.randn(2, 4, 3, 5)
x2 = torch.randn(2, 4, 3, 5)

ARGS = {
    'max_displacement': 4,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Correlation — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a Correlation output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
FlowHead Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("919b54dd2d2863a6")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

HSigmoid Ran
ran on a synthesized input
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("fc85bca2cba2a282")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

class HSigmoid(nn.Module):
    """
    Approximated sigmoid function, so-called hard-version of sigmoid from 'Searching for MobileNetV3,'
    https://arxiv.org/abs/1905.02244.
    """

    def forward(self, x):
        return F.relu6(x + 3.0, inplace=True) / 6.0


INIT_ARGS = {}

ARGS = {
    'x': torch.randn(2, 4)
}

if __name__ == '__main__':
    model = HSigmoid(**INIT_ARGS)
    output = model(**ARGS)
    print(output)
    print(output.shape)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
HSwish Ran
ran on a synthesized input · constructed an object; no result was computed
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("861080d22f42f5ab")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, HSwish: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

torch.manual_seed(0)

ARGS = {
    "inplace": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type HSwish — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a HSwish output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
RAFT Ran
ran on a synthesized input · constructed an object; no result was computed
pytorch/vision/torchvision/models/optical_flow/raft.py
code served (permissive licence) · get_code("0ba6555db6828c81")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned an instance of its own class, RAFT: the run constructed the object and did not call it, so no result was computed. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F

torch.manual_seed(0)

# Build minimal feature encoder
class MinimalFeatureEncoder(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv = nn.Conv2d(6, 16, 3, padding=1)
    
    def forward(self, x):
        # Expects concatenated images, outputs downsampled by 8
        return self.conv(x)

# Build minimal context encoder
class MinimalContextEncoder(nn.Module):
    def __init__(self, hidden_state_size=32):
        super().__init__()
        self.conv = nn.Conv2d(3, hidden_state_size + 8, 3, padding=1)
        self.hidden_state_size = hidden_state_size
    
    def forward(self, x):
        # Expects image1, outputs features that will be split
        return self.conv(x)

# Build minimal correlation block
class MinimalCorrBlock(nn.Module):
    def __init__(self, out_channels=64):
        super().__init__()
        self.out_channels = out_channels
        self._out_channels = out_channels
    
    def build_pyramid(self, feature_map_1, feature_map_2):
        self.fmap1 = feature_map_1
        self.fmap2 = feature_map_2
    
    def index_pyramid(self, coords):
        # coords is the coordinates, return correlation features
        batch_size = coords.shape[0]
        h, w = coords.shape[-2:]
        return torch.randn(batch_size, self.out_channels, h, w)

# Build minimal update block
class MinimalUpdateBlock(nn.Module):
    def __init__(self, hidden_state_size=32, corr_channels=64):
        super().__init__()
        self.hidden_state_size = hidden_state_size
        self.corr_channels = corr_channels
        self.motion_encoder = nn.Conv2d(2, 8, 3, padding=1)
        self.recurrent_unit = nn.LSTMCell(8 + corr_channels + 8, hidden_state_size)
        self.flow_head = nn.Conv2d(hidden_state_size, 2, 3, padding=1)
    
    def forward(self, hidden_state, context, corr_features, flow):
        # Minimal forward pass
        motion = self.motion_encoder(flow)
        cat_input = torch.cat([motion, corr_features, context[:, :8]], dim=1)
        batch_size, _, h, w = cat_input.shape
        cat_input = cat_input.permute(0, 2, 3, 1).reshape(batch_size * h * w, -1)
        if hidden_state is None:
            hidden_state = torch.zeros(batch_size * h * w, self.hidden_state_size)
        hx, cx = self.recurrent_unit(cat_input, (hidden_state, torch.zeros_like(hidden_state)))
        hx = hx.reshape(batch_size, h, w, self.hidden_state_size).permute(0, 3, 1, 2)
        delta_flow = self.flow_head(hx)
        return delta_flow, hx

# Build minimal mask predictor
class MinimalMaskPredictor(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv = nn.Conv2d(32, 9 * 8 * 8, 3, padding=1)
    
    def forward(self, x):
        return self.conv(x)

# Create instances
feature_encoder = MinimalFeatureEncoder()
context_encoder = MinimalContextEncoder(hidden_state_size=32)
corr_block = MinimalCorrBlock(out_channels=64)
update_block = MinimalUpdateBlock(hidden_state_size=32, corr_channels=64)
mask_predictor = MinimalMaskPredictor()

ARGS = {
    'feature_encoder': feature_encoder,
    'context_encoder': context_encoder,
    'corr_block': corr_block,
    'update_block': update_block,
    'mask_predictor': mask_predictor,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type RAFT — the run constructed an instance of its own class and did not call it; no shape is stated — the instrument coerces any output to an array; for a RAFT output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
RSTB Ran
ran on a synthesized input
esakak/sevc/src/models/submodels/RSTB.py
code served (permissive licence) · get_code("be4abbbe7113e5b8")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 5, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.utils.checkpoint as checkpoint
import numpy as np
from timm.models.layers import trunc_normal_

class WindowAttention(nn.Module):
    def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0.0, proj_drop=0.0):
        super().__init__()
        self.dim = dim
        self.window_size = window_size
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** -0.5
        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)
        self.relative_position_bias_table = nn.Parameter(
            torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1) * (2 * window_size[2] - 1), num_heads))
        trunc_normal_(self.relative_position_bias_table, std=.02)

    def forward(self, x, mask=None):
        B, T, H, W, C = x.shape
        qkv = self.qkv(x).reshape(B, T, H, W, 3, self.num_heads, C // self.num_heads).permute(4, 0, 5, 1, 2, 3, 6)
        q, k, v = qkv[0], qkv[1], qkv[2]
        q = q * self.scale
        attn = (q @ k.transpose(-2, -1))
        attn = attn + self.relative_position_bias_table.view(-1, self.num_heads)[None, None, None, None, :]
        if mask is not None:
            nW = mask.shape[0]
            attn = attn.view(B // nW, nW, self.num_heads, T, H, W, T * H * W) + mask.unsqueeze(1).unsqueeze(0)
            attn = attn.view(-1, self.num_heads, T, H, W, T * H * W)
        attn = F.softmax(attn, dim=-1)
        attn = self.attn_drop(attn)
        x = (attn @ v).transpose(1, 2).reshape(B, T, H, W, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x

def drop_path(x, drop_prob=0.0, training=False):
    if drop_prob == 0.0 or not training:
        return x
    keep_prob = 1 - drop_prob
    shape = (x.shape[0],) + (1,) * (x.ndim - 1)
    random_tensor = keep_prob + torch.rand(shape, dtype=x.dtype, device=x.device)
    random_tensor.floor_()
    output = x.div(keep_prob) * random_tensor
    return output

INIT_ARGS = {
    'dim': 4,
    'depth': 2,
    'num_heads': 2,
    'window_size': (2, 3, 4),
    'mlp_ratio': 2.0,
    'qkv_bias': True,
    'qk_scale': None,
    'drop': 0.0,
    'attn_drop': 0.0,
    'drop_path': [0.0, 0.0],
    'norm_layer': nn.LayerNorm,
    'use_checkpoint': False
}

ARGS = {
    'x': torch.randn(2, 4, 6, 8, 4),
    'attn_mask': None
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, timm, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 5, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
ResidualBlock Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("c84b93ca013e3092")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

SepConvGRU Ran
ran on a synthesized input
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("9508f88138c92c66")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

SmallEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("f6d70153459c4013")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

SmallMotionEncoder Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("b70326a63189e916")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

SmallUpdateBlock Ran
ran on a synthesized input · constructed an object; no result was computed
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("b343190f1d5b6e4f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

Swish Ran
ran on a synthesized input
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("6152b289f1a1e079")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

class Swish(nn.Module):
    """
    Swish activation function from 'Searching for Activation Functions,' https://arxiv.org/abs/1710.05941.
    """

    def forward(self, x):
        return x * torch.sigmoid(x)


INIT_ARGS = {}

ARGS = {
    'x': torch.randn(2, 4)
}

if __name__ == '__main__':
    module = Swish(**INIT_ARGS)
    output = module(**ARGS)
    print(output)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
WindowAttention Ran
ran on a synthesized input
esakak/sevc/src/models/submodels/RSTB.py
code served (permissive licence) · get_code("c74d5482daabe793")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 24, 8) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from timm.models.layers import trunc_normal_

class WindowAttention(nn.Module):

    def __init__(self, dim, window_size, num_heads, qkv_bias=True, qk_scale=None, attn_drop=0.0, proj_drop=0.0):
        super().__init__()
        self.dim = dim
        self.window_size = window_size
        self.num_heads = num_heads
        head_dim = dim // num_heads
        self.scale = qk_scale or head_dim ** (-0.5)
        self.relative_position_bias_table = nn.Parameter(torch.zeros((2 * window_size[0] - 1) * (2 * window_size[1] - 1) * (2 * window_size[2] - 1), num_heads))
        coords_d = torch.arange(self.window_size[0])
        coords_h = torch.arange(self.window_size[1])
        coords_w = torch.arange(self.window_size[2])
        coords = torch.stack(torch.meshgrid(coords_d, coords_h, coords_w, indexing='ij'))
        coords_flatten = torch.flatten(coords, 1)
        relative_coords = coords_flatten[:, :, None] - coords_flatten[:, None, :]
        relative_coords = relative_coords.permute(1, 2, 0).contiguous()
        relative_coords[:, :, 0] += self.window_size[0] - 1
        relative_coords[:, :, 1] += self.window_size[1] - 1
        relative_coords[:, :, 2] += self.window_size[2] - 1
        relative_coords[:, :, 0] *= (2 * self.window_size[1] - 1) * (2 * self.window_size[2] - 1)
        relative_coords[:, :, 1] *= 2 * self.window_size[2] - 1
        relative_position_index = relative_coords.sum(-1)
        self.register_buffer('relative_position_index', relative_position_index)
        self.qkv = nn.Linear(dim, dim * 3, bias=qkv_bias)
        self.attn_drop = nn.Dropout(attn_drop)
        self.proj = nn.Linear(dim, dim)
        self.proj_drop = nn.Dropout(proj_drop)
        trunc_normal_(self.relative_position_bias_table, std=0.02)
        self.softmax = nn.Softmax(dim=-1)

    def forward(self, x, mask=None):
        B_, N, C = x.shape
        qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4).contiguous()
        q, k, v = (qkv[0], qkv[1], qkv[2])
        q = q * self.scale
        attn = q @ k.transpose(-2, -1)
        relative_position_bias = self.relative_position_bias_table[self.relative_position_index[:N, :N].reshape(-1)].reshape(N, N, -1)
        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()
        attn = attn + relative_position_bias.unsqueeze(0)
        if mask is not None:
            nW = mask.shape[0]
            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
            attn = attn.view(-1, self.num_heads, N, N)
            attn = self.softmax(attn)
        else:
            attn = self.softmax(attn)
        attn = self.attn_drop(attn)
        x = (attn @ v).transpose(1, 2).reshape(B_, N, C)
        x = self.proj(x)
        x = self.proj_drop(x)
        return x

    def get_attn_Attn(self, x, mask=None):
        B_, N, C = x.shape
        qkv = self.qkv(x).reshape(B_, N, 3, self.num_heads, C // self.num_heads).permute(2, 0, 3, 1, 4).contiguous()
        q, k, v = (qkv[0], qkv[1], qkv[2])
        q = q * self.scale
        attn = q @ k.transpose(-2, -1)
        relative_position_bias = self.relative_position_bias_table[self.relative_position_index[:N, :N].reshape(-1)].reshape(N, N, -1)
        relative_position_bias = relative_position_bias.permute(2, 0, 1).contiguous()
        attn = attn + relative_position_bias.unsqueeze(0)
        if mask is not None:
            nW = mask.shape[0]
            attn = attn.view(B_ // nW, nW, self.num_heads, N, N) + mask.unsqueeze(1).unsqueeze(0)
            attn = attn.view(-1, self.num_heads, N, N)
            attn = self.softmax(attn)
        else:
            attn = self.softmax(attn)
        attn = self.attn_drop(attn)
        return attn


INIT_ARGS = {
    'dim': 8,
    'window_size': (2, 3, 4),
    'num_heads': 2,
    'qkv_bias': True,
    'qk_scale': None,
    'attn_drop': 0.0,
    'proj_drop': 0.0,
}

ARGS = {
    'x': torch.randn(2, 24, 8),
    'mask': None,
}

if __name__ == '__main__':
    model = WindowAttention(**INIT_ARGS)
    output = model(**ARGS)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
timm, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 24, 8) — shape (2, 24, 8) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
bilinear_sampler Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("d5720f5449ed1f1b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 3, 5, 6) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# img: torch.Tensor - should be a 4D tensor (batch, channels, height, width)
# For grid_sample, input should be (N, C, H_in, W_in)
img = torch.randn(2, 3, 8, 10)

# coords: torch.Tensor - coordinates in pixel space
# Based on the function, it splits coords into x_grid and y_grid along dim=-1
# So coords should have shape (..., 2) where last dimension has [x, y] pairs
# For grid_sample compatibility, grid should be (N, H_out, W_out, 2)
# So coords should be (batch_size, height_out, width_out, 2)
coords = torch.rand(2, 5, 6, 2) * torch.tensor([10.0, 8.0])

ARGS = {
    "img": img,
    "coords": coords,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 3, 5, 6) — shape (2, 3, 5, 6) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
conv1x1 Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("ebdde07874936e17")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Conv2d. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "in_channels": 3,
    "out_channels": 64,
    "stride": 1,
    "groups": 1,
    "bias": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Conv2d — a Conv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Conv2d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
conv3x3 Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("7f388bc44b6fceae")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Conv2d. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "in_channels": 3,
    "out_channels": 64,
    "stride": 1,
    "padding": 1,
    "dilation": 1,
    "groups": 1,
    "bias": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Conv2d — a Conv2d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Conv2d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
conv3x3_block Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("9dc286722d0492ff")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned ConvBlock. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Build the minimal valid arguments for conv3x3_block
ARGS = {
    'in_channels': 3,
    'out_channels': 16,
    'stride': 1,
    'padding': 1,
    'dilation': 1,
    'groups': 1,
    'bias': False,
    'normalization': lambda num_features: nn.BatchNorm2d(num_features=num_features, eps=1e-05),
    'activation': lambda: nn.ReLU(inplace=True),
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ConvBlock — a ConvBlock output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ConvBlock output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
coords_grid Ran
ran, honoured the contract we drafted
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("81a137c5322b4930")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

create_activation_layer Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("6ab307b869a70f42")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ReLU. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from typing import Callable

torch.manual_seed(0)

class HSigmoid(nn.Module):
    """
    Approximated sigmoid function, so-called hard-version of sigmoid from 'Searching for MobileNetV3,'
    https://arxiv.org/abs/1905.02244.
    """

    def forward(self, x):
        return torch.nn.functional.relu6(x + 3.0, inplace=True) / 6.0

class HSwish(nn.Module):
    """
    H-Swish activation function from 'Searching for MobileNetV3,' https://arxiv.org/abs/1905.02244.

    Parameters
    ----------
    inplace : bool, default False
        Whether to use inplace version of the module.
    """

    def __init__(self, inplace: bool=False):
        super(HSwish, self).__init__()
        self.inplace = inplace

    def forward(self, x):
        return x * torch.nn.functional.relu6(x + 3.0, inplace=self.inplace) / 6.0

class Swish(nn.Module):
    """
    Swish activation function from 'Searching for Activation Functions,' https://arxiv.org/abs/1710.05941.
    """

    def forward(self, x):
        return x * torch.sigmoid(x)

# Test case 1: string activation
activation_str = 'relu'

# Test case 2: function activation
activation_func = lambda: nn.ReLU(inplace=True)

# Test case 3: nn.Module activation
activation_module = nn.Sigmoid()

ARGS = {
    'activation': activation_str
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ReLU — a ReLU output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a ReLU output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
create_coords_grid Ran
ran, honoured the contract we drafted
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("1ccc23830eea2a4b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 2, 4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    "batch": 2,
    "height": 4,
    "width": 3,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 2, 4, 3) — shape (2, 2, 4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
create_normalization_layer Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("caa1617c550d8b47")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned BatchNorm1d. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn
from typing import Callable

torch.manual_seed(0)

# Test case 1: normalization as a lambda function generator
def normalization_lambda(**kwargs):
    """Lambda function that returns a normalization layer"""
    return nn.BatchNorm1d(num_features=kwargs.get('num_features', 4))

# Test case 2: normalization as an nn.Module instance
normalization_module = nn.BatchNorm1d(num_features=4)

# Create ARGS dict with a lambda function as normalization
ARGS = {
    'normalization': normalization_lambda,
    'num_features': 4
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type BatchNorm1d — a BatchNorm1d output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a BatchNorm1d output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
get_norm_flow Ran
ran on a synthesized input
wpr1018001/rethinking-low-quality-optical-flow/core/flow_aggregation_head_with_residual.py
code served (permissive licence) · get_code("d0c109da15d10c46")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create two flow tensors with shape (batch, channels, height, width)
# batch=2, channels=2, height=4, width=3 (distinct dimensions)
lis1 = torch.randn(2, 2, 4, 3)
lis2 = torch.randn(2, 2, 5, 6)

ARGS = {
    "lis1": lis1,
    "lis2": lis2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple — a tuple output has no array shape of its own; none is stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
initialize_flow Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("ff08a3532e36e78d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 5, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a small image tensor
# Shape: (batch=2, channels=3, height=32, width=40)
img = torch.randn(2, 3, 32, 40)

ARGS = {
    "img": img
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 5, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
lambda_batchnorm2d Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("0aa50f8a83f710b9")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "eps": 1e-05
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
lambda_instancenorm2d Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("91dffaac445c1338")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# The function lambda_instancenorm2d takes eps as a parameter
# and returns a lambda function that creates nn.InstanceNorm2d instances.
# The eps parameter is a float with default 1e-5.

ARGS = {
    "eps": 1e-05
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
lambda_relu Ran
ran; our contract draft was wrong, not the code
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("71d18c7938254ec3")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

ARGS = {
    "inplace": True,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_image Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
wpr1018001/rethinking-low-quality-optical-flow/RAFT/inference_demo.py
code served (permissive licence) · get_code("6c521de8b851930a")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float; no output shape is stated. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import torch
from PIL import Image
import io

# Create a small test image in memory
# Create RGB image (3 channels, 4x5 size)
img_array = np.random.randint(0, 256, (4, 5, 3), dtype=np.uint8)
img = Image.fromarray(img_array, mode='RGB')

# Save to bytes buffer instead of file
img_buffer = io.BytesIO()
img.save(img_buffer, format='PNG')
img_buffer.seek(0)

# Create a temporary file-like object
class FakeImageFile:
    def __init__(self, buffer):
        self.buffer = buffer
    
    def __enter__(self):
        return self.buffer
    
    def __exit__(self, *args):
        pass

# For the function to work with Image.open(), we need an actual file path
# Since we can't do real I/O, we'll create a minimal mock that Image.open can handle
import tempfile
import os

# Create actual temporary file
temp_dir = tempfile.gettempdir()
temp_file = os.path.join(temp_dir, 'test_img_temp.png')
img.save(temp_file)

ARGS = {
    'imfile': temp_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float — no output shape is stated: observe_signature.py could not observe it (call:AssertionError)
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
make_coords_grid Ran
ran, honoured the contract we drafted
pytorch/vision/torchvision/models/optical_flow/raft.py
code served (permissive licence) · get_code("a7aae9f6f384f21e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 2, 4, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

ARGS = {
    'batch_size': 2,
    'h': 4,
    'w': 3,
    'device': 'cpu'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 2, 4, 3) — shape (2, 2, 4, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
upflow8 Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("561d1e8d3102a282")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

upsample_flow Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
pytorch/vision/torchvision/models/optical_flow/raft.py
code served (permissive licence) · get_code("f265d59c1942f6bb")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 2, 32, 24) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Create a small flow tensor: batch_size=2, num_channels=2, h=4, w=3
flow = torch.randn(2, 2, 4, 3)

# Create up_mask tensor with shape (batch_size, 1, 9, factor, factor, h, w)
# where factor=8, so shape is (2, 1, 9, 8, 8, 4, 3)
up_mask = torch.randn(2, 1, 9, 8, 8, 4, 3)

factor = 8

ARGS = {
    "flow": flow,
    "up_mask": up_mask,
    "factor": factor
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 2, 32, 24) — shape (2, 2, 32, 24) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
upsample_flow_using_mask Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("f70dd0be2dffde5d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 4, dtype family float, shape (2, 2, 32, 24) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn.functional as F

torch.manual_seed(0)

# Create a small flow tensor with shape [batch=2, channels=2, height=4, width=3]
# The function expects [batch, 2, H/8, W/8], so we use small H and W
flow = torch.randn(2, 2, 4, 3)

# Create a mask tensor
# The function reshapes mask to [batch, 1, 9, 8, 8, height, width]
# Working backwards: mask.view(batch, 1, 9, 8, 8, height, width)
# So the input mask should have total elements = batch * 1 * 9 * 8 * 8 * height * width
# = 2 * 1 * 9 * 8 * 8 * 4 * 3 = 17280
# We can create it as [batch, 9*64, height, width] = [2, 576, 4, 3]
mask = torch.randn(2, 576, 4, 3)

ARGS = {
    "flow": flow,
    "mask": mask,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 4, dtype family float, shape (2, 2, 32, 24) — shape (2, 2, 32, 24) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
window_partition Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
esakak/sevc/src/models/submodels/RSTB.py
code served (permissive licence) · get_code("5f65d2c7face3e5d")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 3, dtype family float, shape (24, 16, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# The function expects:
# - x: a 5D tensor with shape (B, D, H, W, C)
# - window_size: a sequence of 3 integers for partitioning dimensions D, H, W

# Create a small 5D tensor
# B=2 (batch), D=4 (depth), H=6 (height), W=8 (width), C=3 (channels)
x = torch.randn(2, 4, 6, 8, 3)

# window_size must divide evenly into D, H, W
# D=4 divisible by 2, H=6 divisible by 2, W=8 divisible by 2
window_size = (2, 2, 4)

ARGS = {
    "x": x,
    "window_size": window_size
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (24, 16, 3) — shape (24, 16, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
window_reverse Ran
ran; our contract draft was wrong, not the code
esakak/sevc/src/models/submodels/RSTB.py
code served (permissive licence) · get_code("6faaa9eebd9890d7")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 5, dtype family float, shape (2, 8, 6, 4, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import torch.nn as nn

torch.manual_seed(0)

# Analyze the function:
# - windows: tensor that gets reshaped and permuted
# - window_size: tuple/list with 3 elements (indexing [0], [1], [2])
# - B: batch size (int)
# - D: depth dimension (int)
# - H: height dimension (int)
# - W: width dimension (int)

# The function reshapes windows from shape (B*num_windows, window_size[0]*window_size[1]*window_size[2], C)
# to (B, D, H, W, C)

# Working backwards from the view operations:
# Final shape should be (B, D, H, W, -1)
# Before final view: (B, D//window_size[0], H//window_size[1], W//window_size[2], window_size[0], window_size[1], window_size[2], -1)
# So windows input shape should be compatible with this reshape

B = 2
D = 8
H = 6
W = 4
window_size = (2, 3, 2)  # (depth_window, height_window, width_window)
C = 5  # channel dimension

# windows shape should be:
# (B * (D//window_size[0]) * (H//window_size[1]) * (W//window_size[2]), window_size[0]*window_size[1]*window_size[2], C)
num_windows_d = D // window_size[0]  # 8 // 2 = 4
num_windows_h = H // window_size[1]  # 6 // 3 = 2
num_windows_w = W // window_size[2]  # 4 // 2 = 2
total_windows = num_windows_d * num_windows_h * num_windows_w  # 4 * 2 * 2 = 16

windows_shape = (B * total_windows, window_size[0] * window_size[1] * window_size[2], C)
windows = torch.randn(windows_shape)

ARGS = {
    "windows": windows,
    "window_size": window_size,
    "B": B,
    "D": D,
    "H": H,
    "W": W
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 5, dtype family float, shape (2, 8, 6, 4, 5) — shape (2, 8, 6, 4, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
AltCudaCorr Not yet run
unverified (harvested, no recorded run)
princeton-vl/RAFT/core/raft.py
code served (permissive licence) · get_code("4cb42bca3a610fbe")
How it ran

no invocation recorded.

AlternateCorrBlock Not yet run
unverified (harvested, no recorded run)
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("de07f7177a45e51f")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

AlternateCorrBlock Not yet run
unverified (harvested, no recorded run)
princeton-vl/RAFT/core/raft.py
code served (permissive licence) · get_code("5dd835869366517d")
How it ran

no invocation recorded.

BasicLayer Not yet run
unverified (harvested, no recorded run)
esakak/sevc/src/models/submodels/RSTB.py
code served (permissive licence) · get_code("6e27a7978c935cc6")
How it ran

no invocation recorded.

BasicUpdateBlock Not yet run
unverified (harvested, no recorded run)
aliciachenw/PIPsUS/nets/update.py
code served (permissive licence) · get_code("f8f994317c0ac6e7")
How it ran

no invocation recorded.

ConvGRU Not yet run
unverified (harvested, no recorded run)
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("1ecbeae1d13a0815")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

ConvSeqBranch Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("876f851f8a5dab7a")
How it ran

no invocation recorded.

FlowHead Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("576170c699910f7c")
How it ran

no invocation recorded.

MaskHead Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("1fd5a8b38dc17f33")
How it ran

no invocation recorded.

PatchEmbed Not yet run
unverified (harvested, no recorded run)
esakak/sevc/src/models/submodels/RSTB.py
code served (permissive licence) · get_code("f0e19d80e6ff4884")
How it ran

no invocation recorded.

PatchUnEmbed Not yet run
unverified (harvested, no recorded run)
esakak/sevc/src/models/submodels/RSTB.py
code served (permissive licence) · get_code("7d0bab782fa467e4")
How it ran

no invocation recorded.

RAFT Not yet run
unverified (harvested, no recorded run)
plusgood-steven/id-blau/PrepareCondition/core/raft.py
pointer only (licence: NONE) · get_code("c9bcaa550ff44555")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, plusgood-steven/id-blau when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

RAFT Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("bd49469ee4db2086")
How it ran

no invocation recorded.

RAFT Not yet run
unverified (harvested, no recorded run)
princeton-vl/RAFT/core/raft.py
code served (permissive licence) · get_code("cb380d6b60e23020")
How it ran

no invocation recorded.

RAFTEncoder Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("12dcf7f00f5c8dac")
How it ran

no invocation recorded.

RAFTMotionEncoder Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("88471cb2aa391635")
How it ran

no invocation recorded.

RAFTUpdateBlock Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("9376ca36060a23ee")
How it ran

no invocation recorded.

ResBlock Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("8f9a47ae470ac10f")
How it ran

no invocation recorded.

ResBottleneck Not yet run
unverified (harvested, no recorded run)
osmr/pytorchcv/pytorchcv/models/raft.py
code served (permissive licence) · get_code("89f3a46e1c990919")
How it ran

no invocation recorded.

Showing the first 60. An agent gets every one from get_harvested_code_for_paper("2003.12039").

Also present in other papers' code

Of 66 samples lifted from this paper's linked repositories, 66 carry a function identity and 0 do not. Of the 64 distinct functions with an identity, 30 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 30 later) and 34 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
bilinear_sampler
brick of this paper's assembly: helper
18 other papers
18 later · 7 only as an assembly brick
2108.00622 (2021-08)
Recurrent Mask Refinement for Few-Shot Medical Image Segmentation
2108.10552 (2021-08)
E-RAFT: Dense Optical Flow from Event Cameras
2109.15321 (2021-09)
Sensor-Guided Optical Flow
and 15 more not named here
window_reverse
brick of this paper's assembly: helper
16 other papers
16 later · 5 only as an assembly brick
2106.13230 (2021-06)
Video Swin Transformer
2111.14821 (2021-11)
End-to-End Referring Video Object Segmentation with Multimodal Transformers
2207.02595 (2022-07)
FAST-VQA: Efficient End-to-end Video Quality Assessment with Fragment Sampling
and 13 more not named here
Swish
brick of this paper's assembly: helper
10 other papers
3 dated earlier, 7 later · 7 only as an assembly brick
1505.04597 (2015-05)
U-Net: Convolutional Networks for Biomedical Image Segmentation
1612.03144 (2016-12)
Feature Pyramid Networks for Object Detection
1905.11946 (2019-05)
EfficientNet: Rethinking Model Scaling for Convolutional Neural Networks
and 7 more not named here
SepConvGRU
brick of this paper's assembly: model, update_rule
8 other papers
8 later · 8 only as an assembly brick
2202.03857 (2022-02)
Learning Optical Flow with Adaptive Graph Reasoning
2203.11335 (2022-03)
Global Matching with Overlapping Attention for Optical Flow Estimation
2203.11483 (2022-03)
Practical Stereo Matching via Cascaded Recurrent Network with Adaptive Correlation
and 5 more not named here
coords_grid
brick of this paper's assembly: helper
7 other papers
7 later · 5 only as an assembly brick
2202.03857 (2022-02)
Learning Optical Flow with Adaptive Graph Reasoning
2308.13133 (2023-08)
AccFlow: Backward Accumulation for Long-Range Optical Flow
2405.14793 (2024-05)
SEA-RAFT: Simple, Efficient, Accurate RAFT for Optical Flow
and 4 more not named here
FlowHead
brick of this paper's assembly: model
6 other papers
6 later · 6 only as an assembly brick
2203.11335 (2022-03)
Global Matching with Overlapping Attention for Optical Flow Estimation
2203.11483 (2022-03)
Practical Stereo Matching via Cascaded Recurrent Network with Adaptive Correlation
2204.00330 (2022-04)
DIP: Deep Inverse Patchmatch for High-Resolution Optical Flow
and 3 more not named here
ResidualBlock
brick of this paper's assembly: helper
6 other papers
6 later · 6 only as an assembly brick
2202.03857 (2022-02)
Learning Optical Flow with Adaptive Graph Reasoning
2203.11335 (2022-03)
Global Matching with Overlapping Attention for Optical Flow Estimation
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
and 3 more not named here
upflow8
brick of this paper's assembly: helper
5 other papers
5 later · 4 only as an assembly brick
2108.10552 (2021-08)
E-RAFT: Dense Optical Flow from Event Cameras
2202.03857 (2022-02)
Learning Optical Flow with Adaptive Graph Reasoning
2203.11335 (2022-03)
Global Matching with Overlapping Attention for Optical Flow Estimation
and 2 more not named here
AlternateCorrBlock
brick of this paper's assembly: model
4 other papers
4 later · 4 only as an assembly brick
2202.03857 (2022-02)
Learning Optical Flow with Adaptive Graph Reasoning
2203.11335 (2022-03)
Global Matching with Overlapping Attention for Optical Flow Estimation
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
and 1 more not named here
ConvGRU
brick of this paper's assembly: model
4 other papers
4 later · 3 only as an assembly brick
2204.00330 (2022-04)
DIP: Deep Inverse Patchmatch for High-Resolution Optical Flow
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
2406.13457 (2024-06)
EvTexture: Event-driven Texture Enhancement for Video Super-Resolution
and 1 more not named here
load_image 4 other papers
1 dated earlier, 3 later
1512.03385 (2015-12)
Deep Residual Learning for Image Recognition
2401.04861 (2024-01)
CTNeRF: Cross-Time Transformer for Dynamic Neural Radiance Field from Monocular Video
2404.05163 (2024-04)
SEMANTIC FLOW: LEARNING SEMANTIC FIELDS OF DYNAMIC SCENES FROM MONOCULAR VIDEOS
and 1 more not named here
BasicEncoder
brick of this paper's assembly: model
3 other papers
3 later · 3 only as an assembly brick
2202.03857 (2022-02)
Learning Optical Flow with Adaptive Graph Reasoning
2203.11335 (2022-03)
Global Matching with Overlapping Attention for Optical Flow Estimation
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
BottleneckBlock
brick of this paper's assembly: helper
3 other papers
3 later · 3 only as an assembly brick
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
2304.09790 (2023-04)
AMT: All-Pairs Multi-Field Transforms for Efficient Frame Interpolation
2507.23595 (2025-07)
MamV2XCalib: V2X-based Target-less Infrastructure Camera Calibration with State Space Model
CorrBlock
brick of this paper's assembly: model
3 other papers
3 later · 2 only as an assembly brick
2202.03857 (2022-02)
Learning Optical Flow with Adaptive Graph Reasoning
2406.00429 (2024-06)
Towards Generalizable Multi-Object Tracking
2507.23595 (2025-07)
MamV2XCalib: V2X-based Target-less Infrastructure Camera Calibration with State Space Model
SmallEncoder
brick of this paper's assembly: model
3 other papers
3 later · 3 only as an assembly brick
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
2304.09790 (2023-04)
AMT: All-Pairs Multi-Field Transforms for Efficient Frame Interpolation
2507.23595 (2025-07)
MamV2XCalib: V2X-based Target-less Infrastructure Camera Calibration with State Space Model
BasicMotionEncoder
brick of this paper's assembly: model
2 other papers
2 later · 2 only as an assembly brick
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
2506.23897 (2025-06)
PriOr-Flow: Enhancing Primitive Panoramic Optical Flow with Orthogonal View
BasicUpdateBlock
brick of this paper's assembly: entry, update_rule
2 other papers
2 later · 2 only as an assembly brick
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
2506.23897 (2025-06)
PriOr-Flow: Enhancing Primitive Panoramic Optical Flow with Orthogonal View
SmallMotionEncoder
brick of this paper's assembly: model
2 other papers
2 later · 2 only as an assembly brick
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
2507.23595 (2025-07)
MamV2XCalib: V2X-based Target-less Infrastructure Camera Calibration with State Space Model
SmallUpdateBlock
brick of this paper's assembly: update_rule
2 other papers
2 later · 2 only as an assembly brick
2207.03790 (2022-07)
Complementing Brightness Constancy with Deep Networks for Optical Flow Prediction
2507.23595 (2025-07)
MamV2XCalib: V2X-based Target-less Infrastructure Camera Calibration with State Space Model
window_partition
brick of this paper's assembly: helper
2 other papers
2 later
2206.02146 (2022-06)
Recurrent Video Restoration Transformer with Guided Deformable Attention
2305.00163 (2023-05)
Enhancing Video Super-Resolution via Implicit Resampling-based Alignment

Shown: the 20 functions present in the most other papers. 10 more shared functions are not listed here.

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-22.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

We introduce Recurrent All-Pairs Field Transforms (RAFT), a new deep network architecture for optical flow. RAFT extracts perpixel features, builds multi-scale 4D correlation volumes for all pairs of pixels, and iteratively updates a flow field through a recurrent unit that performs lookups on the correlation volumes. RAFT achieves stateof-the-art performance. On KITTI, RAFT achieves an F1-all error of 5.10%, a 16% error reduction from the best published result (6.10%). On Sintel (final pass), RAFT obtains an end-point-error of 2.855 pixels, a 30% error reduction from the best published result (4.098 pixels). In addition, RAFT has strong cross-dataset generalization as well as high efficiency in inference time, training speed, and parameter count. Code is available at https://github.com/princeton-vl/RAFT.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2003.12039")
get_code_for_paper("2003.12039")
have("2003.12039")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2003.12039.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2003.12039.svg)](https://syntology.ai/paper/2003.12039)

Connect an agent — have() is free.