SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2207.02206 · NeurIPS · 2022

Segmenting Moving Objects via an Object-Centric Layered Representation

Andrew Zisserman, Junyu Xie, Weidi Xie

arXiv · PDF · Open in the Atlas

Code that ran

16 of 21 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (9 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 11 have no contract verdict; 4 failures of our instrument; 5 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/2207.02206.json · badge: /api/ran/2207.02206.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
Jyxarthur/OCLR_model marked official in Papers with Code
Papers with Code archive
16 ran · 5 no recorded run
FunctionStatusWhere it lives
AttnMap Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("867da52dc8c76b0f")
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, AttnMap: 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 einops
import torch.nn as nn

torch.manual_seed(0)

class AttnMap(nn.Module):

    def __init__(self, btn_features, num_heads, num_query):
        super(AttnMap, self).__init__()
        self.heads = num_heads
        self.mlp = nn.Sequential(nn.Linear(num_heads * num_query, btn_features), nn.ReLU(inplace=True))

    def forward(self, btn_dec, btn_enc):
        b, t, h, w, _ = btn_enc.size()
        btn_enc = einops.rearrange(btn_enc, 'b t h w (g c) -> (b g) t h w c', g=self.heads)
        btn_dec = einops.rearrange(btn_dec, 'b (q t) (g c) -> (b g) q t c', t=t, g=self.heads)
        btn_out = btn_dec[:, :, :, None, None, :] * btn_enc[:, None, :, :, :, :]
        btn_out = torch.sum(btn_out, 5)
        btn_out = einops.rearrange(btn_out, '(b g) q t h w -> b (t h w) (q g)', g=self.heads)
        btn_out = self.mlp(btn_out)
        btn_out = einops.rearrange(btn_out, 'b (t h w) c -> b t h w c', t=t, h=h, w=w)
        return btn_out

# Set up parameters
btn_features = 16
num_heads = 2
num_query = 3

# Batch size
b = 2
# Time dimension
t = 2
# Height and width
h = 3
w = 4
# Channel dimension (must be divisible by num_heads)
c = 8

# btn_enc shape: (b, t, h, w, c) where c = num_heads * c_per_head
btn_enc = torch.randn(b, t, h, w, c)

# btn_dec shape: (b, q*t, num_heads*c_per_head)
btn_dec = torch.randn(b, num_query * t, num_heads * (c // num_heads))

ARGS = {
    'btn_features': btn_features,
    'num_heads': num_heads,
    'num_query': num_query,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type AttnMap — 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 AttnMap 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
ConvBlock Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("75d889c61752310c")
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
from collections import OrderedDict

torch.manual_seed(0)

ARGS = {
    'in_channels': 3,
    'features': 64,
    'name': 'enc_'
}
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
EncoderLayer Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("8378236c795fdc48")
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, EncoderLayer: 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)

ARGS = {
    "d_model": 4,
    "heads": 2,
    "dim_feedforward": 8,
    "dropout": 0.1,
}
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 EncoderLayer — 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 EncoderLayer 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
MultiHeadAttention Ran
ran on a synthesized input
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("a447981fffd42b39")
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; 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 math
import torch.nn.functional as F
from typing import List, Optional
from torch.nn.parameter import Parameter
from torch.nn.init import xavier_uniform_

Tensor = torch.Tensor

def _in_projection_packed(q: Tensor, k: Tensor, v: Tensor, w: Tensor, b: Optional[Tensor]=None) -> List[Tensor]:
    E = q.size(-1)
    if k is v:
        if q is k:
            return F.linear(q, w, b).chunk(3, dim=-1)
        else:
            w_q, w_kv = w.split([E, E * 2])
            if b is None:
                b_q = b_kv = None
            else:
                b_q, b_kv = b.split([E, E * 2])
            return (F.linear(q, w_q, b_q),) + F.linear(k, w_kv, b_kv).chunk(2, dim=-1)
    else:
        w_q, w_k, w_v = w.chunk(3)
        if b is None:
            b_q = b_k = b_v = None
        else:
            b_q, b_k, b_v = b.chunk(3)
        return (F.linear(q, w_q, b_q), F.linear(k, w_k, b_k), F.linear(v, w_v, b_v))

def attention(q, k, v, d_k, mask=None, dropout=None):
    scores = torch.matmul(q, k.transpose(-2, -1)) / math.sqrt(d_k)
    if mask is not None:
        mask = mask.unsqueeze(0).unsqueeze(0)
        scores = scores.masked_fill(mask == 1, -1000000000.0)
    scores = F.softmax(scores, dim=-1)
    if dropout is not None:
        scores = dropout(scores)
    output = torch.matmul(scores, v)
    return output

class MultiHeadAttention(nn.Module):

    def __init__(self, heads, d_model, dropout=0.1):
        super().__init__()
        self.d_model = d_model
        self.d_k = d_model // heads
        self.h = heads
        self.in_proj_weight = Parameter(torch.empty(3 * d_model, d_model))
        xavier_uniform_(self.in_proj_weight)
        self.in_proj_bias = Parameter(torch.zeros(3 * d_model))
        self.dropout = nn.Dropout(dropout)
        self.out_proj = nn.Linear(d_model, d_model)

    def forward(self, q, k, v, mask=None):
        bs = q.size(0)
        q, k, v = _in_projection_packed(q, k, v, self.in_proj_weight, self.in_proj_bias)
        k = k.view(bs, -1, self.h, self.d_k)
        q = q.view(bs, -1, self.h, self.d_k)
        v = v.view(bs, -1, self.h, self.d_k)
        k = k.transpose(1, 2)
        q = q.transpose(1, 2)
        v = v.transpose(1, 2)
        scores = attention(q, k, v, self.d_k, mask, self.dropout)
        concat = scores.transpose(1, 2).contiguous().view(bs, -1, self.d_model)
        output = self.out_proj(concat)
        return output

INIT_ARGS = {
    'heads': 2,
    'd_model': 8,
    'dropout': 0.1
}

ARGS = {
    'q': torch.randn(2, 5, 8),
    'k': torch.randn(2, 7, 8),
    'v': torch.randn(2, 7, 8),
    'mask': None
}
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 3, 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)
Norm Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("55e3d071afe4d46b")
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, Norm: 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)

# The Norm class takes d_model and eps as constructor parameters
# d_model: the dimension of the model (size of weight and bias parameters)
# eps: small value for numerical stability in normalization

# Create a small instance with d_model=4 (matching our tensor dimensions below)
d_model = 4
eps = 1e-5

# The ARGS dict maps parameter names to their values
ARGS = {
    'd_model': d_model,
    'eps': eps
}
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 Norm — 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 Norm 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
OrderingHead Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("f66e673c9146f36d")
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, OrderingHead: 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 einops
import torch.nn as nn

torch.manual_seed(0)

class OrderingHead(nn.Module):

    def __init__(self, btn_features, num_query):
        super(OrderingHead, self).__init__()
        self.num_query = num_query
        self.mlp = nn.Sequential(nn.Linear(btn_features, btn_features), nn.ReLU(inplace=True), nn.Linear(btn_features, 1))

    def forward(self, btn_dec):
        btn_dec = einops.rearrange(btn_dec, 'b (q t) c -> (b q) t c', q=self.num_query)
        btn_order = torch.max(btn_dec, 1)[0]
        btn_order = self.mlp(btn_order)
        btn_order = einops.rearrange(btn_order, '(b q) c -> b q c', q=self.num_query)
        return btn_order.squeeze(2)

# Create instance for testing forward pass
btn_features = 8
num_query = 3
model = OrderingHead(btn_features=btn_features, num_query=num_query)

# Create valid input tensor
# Shape needs to be: b (q t) c where b=batch, q=num_query, t=sequence length, c=channels
# Let's use: batch=2, q=3, t=4, c=8
# So btn_dec shape should be (2, 12, 8) where 12 = 3*4
batch_size = 2
sequence_length = 4
btn_dec = torch.randn(batch_size, num_query * sequence_length, btn_features)

ARGS = {
    "btn_features": btn_features,
    "num_query": num_query
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type OrderingHead — 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 OrderingHead 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
SoftPositionTimeEmbed Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("47038bfa9f36a207")
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, SoftPositionTimeEmbed: the run constructed the object and did not call it, so no result was computed. 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 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 numpy as np

torch.manual_seed(0)

def build_spacetime_grid(frames, resolution):
    device = 'cuda' if torch.cuda.is_available() else 'cpu'
    ranges = [np.linspace(-1.0, 1.0, num=frames)] + [np.linspace(-1.0, 1.0, num=res) for res in resolution]
    grid = np.meshgrid(*ranges, sparse=False, indexing='ij')
    grid = np.stack(grid, axis=-1)
    grid = np.reshape(grid, [frames, resolution[0], resolution[1], -1])
    grid = np.expand_dims(grid, axis=0)
    grid = grid.astype(np.float32)
    return torch.from_numpy(grid).to(device)

class SoftPositionTimeEmbed(nn.Module):

    def __init__(self, hidden_size):
        super(SoftPositionTimeEmbed, self).__init__()
        self.proj = nn.Linear(3, hidden_size)
        self.grid = build_spacetime_grid(5, (8, 14))

    def forward(self, inputs, frames, resolution):
        if self.grid.size()[1:4] != (frames, resolution[0], resolution[1]):
            self.grid = build_spacetime_grid(frames, resolution)
        return inputs + self.proj(self.grid)

ARGS = {
    'hidden_size': 4
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SoftPositionTimeEmbed — 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 SoftPositionTimeEmbed 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
SoftTimeEmbed Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("fcbab3ccaf3b573d")
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, SoftTimeEmbed: the run constructed the object and did not call it, so no result was computed. 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 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 numpy as np

torch.manual_seed(0)

ARGS = {
    'hidden_size': 8
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
einops, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type SoftTimeEmbed — 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 SoftTimeEmbed 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
TransDecoder Ran
ran on a synthesized input
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("d4e6b98e360c24bf")
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; 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 numpy as np

INIT_ARGS = {
    "d_model": 4,
    "num_layers": 2,
    "nhead": 2,
    "dim_feedforward": 8,
}

# Create input tensors for forward pass
batch_size = 2
seq_len = 3
d_model = INIT_ARGS["d_model"]

x = torch.randn(batch_size, seq_len, d_model)  # target input
e_outputs = torch.randn(batch_size, seq_len, d_model)  # encoder outputs

# Create masks (binary, 1 = masked)
src_mask = torch.zeros(batch_size, 1, seq_len)  # source mask (encoder)
trg_mask = torch.triu(torch.ones(seq_len, seq_len), diagonal=1).bool()  # target mask (future positions)

ARGS = {
    "x": x,
    "e_outputs": e_outputs,
    "src_mask": src_mask,
    "trg_mask": trg_mask,
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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 — 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)
UNet_decoder Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("1c0dce82d6701d1a")
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, UNet_decoder: 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
from collections import OrderedDict

torch.manual_seed(0)

class ConvBlock(nn.Module):
    """ Double convolution block used in U-Net"""

    def __init__(self, in_channels, features, name):
        super(ConvBlock, self).__init__()
        self.conv_block = nn.Sequential(OrderedDict([(name + 'conv1', nn.Conv2d(in_channels=in_channels, out_channels=features, kernel_size=3, padding=1, bias=False)), (name + 'norm1', nn.InstanceNorm2d(num_features=features)), (name + 'relu1', nn.LeakyReLU(inplace=True)), (name + 'conv2', nn.Conv2d(in_channels=features, out_channels=features, kernel_size=3, padding=1, bias=False)), (name + 'norm2', nn.InstanceNorm2d(num_features=features)), (name + 'relu2', nn.LeakyReLU(inplace=True))]))

    def forward(self, x):
        return self.conv_block(x)

class UNet_decoder(nn.Module):
    """ input dimension: b c h w (batch_size channel height width)"""
    ' output dimension: b c H W'

    def __init__(self, out_channels, init_features):
        super(UNet_decoder, self).__init__()
        features = init_features
        self.upconv4 = nn.ConvTranspose2d(features * 16, features * 8, kernel_size=2, stride=2)
        self.decoder4 = ConvBlock(features * 8 * 2, features * 8, name='dec4')
        self.upconv3 = nn.ConvTranspose2d(features * 8, features * 4, kernel_size=2, stride=2)
        self.decoder3 = ConvBlock(features * 4 * 2, features * 4, name='dec3')
        self.upconv2 = nn.ConvTranspose2d(features * 4, features * 2, kernel_size=2, stride=2)
        self.decoder2 = ConvBlock(features * 2 * 2, features * 2, name='dec2')
        self.upconv1 = nn.ConvTranspose2d(features * 2, features, kernel_size=2, stride=2)
        self.decoder1 = ConvBlock(features * 2, features, name='dec1')
        self.conv = nn.Conv2d(in_channels=features, out_channels=out_channels, kernel_size=1)

    def forward(self, enc1, enc2, enc3, enc4, bottleneck):
        dec4 = self.upconv4(bottleneck)
        dec4 = torch.cat((dec4, enc4), dim=1)
        dec4 = self.decoder4(dec4)
        dec3 = self.upconv3(dec4)
        dec3 = torch.cat((dec3, enc3), dim=1)
        dec3 = self.decoder3(dec3)
        dec2 = self.upconv2(dec3)
        dec2 = torch.cat((dec2, enc2), dim=1)
        dec2 = self.decoder2(dec2)
        dec1 = self.upconv1(dec2)
        dec1 = torch.cat((dec1, enc1), dim=1)
        dec1 = self.decoder1(dec1)
        return self.conv(dec1)


# Build minimal tensors for the forward pass
# Batch size: 2, init_features: 1
# Working backwards from output dimension
# Output should be at full resolution (16x16)
# enc1, enc2, enc3, enc4 are at different scales
batch_size = 2
init_features = 1

# For upconv4: input is (batch, features*16, 2, 2) -> output (batch, features*8, 4, 4)
bottleneck = torch.randn(batch_size, init_features * 16, 2, 2)

# enc4: (batch, features*8, 4, 4)
enc4 = torch.randn(batch_size, init_features * 8, 4, 4)

# After upconv3: (batch, features*4, 8, 8)
enc3 = torch.randn(batch_size, init_features * 4, 8, 8)

# After upconv2: (batch, features*2, 16, 16)
enc2 = torch.randn(batch_size, init_features * 2, 16, 16)

# After upconv1: (batch, features, 32, 32)
enc1 = torch.randn(batch_size, init_features, 32, 32)

ARGS = {
    'out_channels': 3,
    'init_features': 1,
}
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 UNet_decoder — 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 UNet_decoder 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
UNet_encoder Ran
ran on a synthesized input · constructed an object; no result was computed
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("5a52ccb38fd0d98e")
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, UNet_encoder: 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
from collections import OrderedDict

torch.manual_seed(0)

class ConvBlock(nn.Module):
    """ Double convolution block used in U-Net"""

    def __init__(self, in_channels, features, name):
        super(ConvBlock, self).__init__()
        self.conv_block = nn.Sequential(OrderedDict([(name + 'conv1', nn.Conv2d(in_channels=in_channels, out_channels=features, kernel_size=3, padding=1, bias=False)), (name + 'norm1', nn.InstanceNorm2d(num_features=features)), (name + 'relu1', nn.LeakyReLU(inplace=True)), (name + 'conv2', nn.Conv2d(in_channels=features, out_channels=features, kernel_size=3, padding=1, bias=False)), (name + 'norm2', nn.InstanceNorm2d(num_features=features)), (name + 'relu2', nn.LeakyReLU(inplace=True))]))

    def forward(self, x):
        return self.conv_block(x)

class UNet_encoder(nn.Module):
    """ input dimension: b c H W (batch_size channel height width)"""
    ' output dimension: b c h w '

    def __init__(self, in_channels, init_features):
        super(UNet_encoder, self).__init__()
        features = init_features
        self.encoder1 = ConvBlock(in_channels, features, name='enc1')
        self.pool1 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.encoder2 = ConvBlock(features, features * 2, name='enc2')
        self.pool2 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.encoder3 = ConvBlock(features * 2, features * 4, name='enc3')
        self.pool3 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.encoder4 = ConvBlock(features * 4, features * 8, name='enc4')
        self.pool4 = nn.MaxPool2d(kernel_size=2, stride=2)
        self.bottleneck = ConvBlock(features * 8, features * 16, name='bottleneck')

    def forward(self, x):
        enc1 = self.encoder1(x)
        enc2 = self.encoder2(self.pool1(enc1))
        enc3 = self.encoder3(self.pool2(enc2))
        enc4 = self.encoder4(self.pool3(enc3))
        bottleneck = self.bottleneck(self.pool4(enc4))
        return (enc1, enc2, enc3, enc4, bottleneck)

ARGS = {
    'in_channels': 3,
    'init_features': 8
}
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 UNet_encoder — 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 UNet_encoder 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
_in_projection_packed Ran
ran; our contract draft was wrong, not the code
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("a6a5c2077b398c5d")
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 4, 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

torch.manual_seed(0)

# Analyze the function:
# - q, k, v are Tensors used with F.linear and .size(-1)
# - w is a Tensor used with F.linear, .split(), and .chunk()
# - b is Optional[Tensor] used with .split() if not None

# The function performs linear projections for multi-head attention
# E = q.size(-1) is the embedding dimension
# w should have shape (3*E, E) for the packed case where q, k, v are different
# b should have shape (3*E,) if provided

# Set embedding dimension
embed_dim = 4
batch_size = 2
seq_len = 3

# Create q, k, v tensors
# q: (batch_size, seq_len, embed_dim)
q = torch.randn(batch_size, seq_len, embed_dim)
k = torch.randn(batch_size, seq_len, embed_dim)
v = torch.randn(batch_size, seq_len, embed_dim)

# Create weight matrix w
# For the case where q, k, v are all different, w should be (3*embed_dim, embed_dim)
w = torch.randn(3 * embed_dim, embed_dim)

# Create optional bias b
# For the case where q, k, v are all different, b should be (3*embed_dim,)
b = torch.randn(3 * embed_dim)

ARGS = {
    "q": q,
    "k": k,
    "v": v,
    "w": w,
    "b": b,
}
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 4, 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
attention Ran
ran; our contract draft was wrong, not the code
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("86341ad6b5c2c7a4")
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 6, dtype family float, shape (1, 1, 2, 1, 3, 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 math

torch.manual_seed(0)

# Construct tiny tensors for multi-head attention
batch_size = 2
seq_len = 3
d_k = 4
num_heads = 1

# q, k, v: (batch_size, num_heads, seq_len, d_k)
q = torch.randn(batch_size, num_heads, seq_len, d_k)
k = torch.randn(batch_size, num_heads, seq_len, d_k)
v = torch.randn(batch_size, num_heads, seq_len, d_k)

# mask: (batch_size, 1, seq_len, seq_len) with 0s and 1s
# 1 indicates positions to mask (attend to nothing)
mask = torch.zeros(batch_size, 1, seq_len, seq_len, dtype=torch.long)
mask[0, 0, 2, :] = 1  # mask out last position in first batch

# dropout layer
dropout = nn.Dropout(p=0.1)

ARGS = {
    "q": q,
    "k": k,
    "v": v,
    "d_k": d_k,
    "mask": mask,
    "dropout": dropout,
}
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 6, dtype family float, shape (1, 1, 2, 1, 3, 4) — shape (1, 1, 2, 1, 3, 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: none by the harness; any seed is in the setup block
attn_mask Ran
ran; our contract draft was wrong, not the code
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("6cb4b60cb90aeb1d")
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 2, dtype family float, shape (24, 24) as a later run recorded it. 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, 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 = {
    'frames': 2,
    'resolution': (4, 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 2, dtype family float, shape (24, 24) — shape (24, 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
build_spacetime_grid Ran
ran; our contract draft was wrong, not the code
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("584cc60931efa04b")
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 5, dtype family float, shape (1, 5, 4, 3, 3) as a later run recorded it. 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, 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)

ARGS = {
    "frames": 5,
    "resolution": [4, 3]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
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 5, dtype family float, shape (1, 5, 4, 3, 3) — shape (1, 5, 4, 3, 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
build_time_grid Ran
ran, honoured the contract we drafted
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("7a79f5c1eb6acc1b")
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 3, dtype family float, shape (1, 16, 1) as a later run recorded it. 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, 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)

ARGS = {
    'frames': 16
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
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 (1, 16, 1) — shape (1, 16, 1) 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
BTNTransformerDec Not yet run
unverified (harvested, no recorded run)
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("34a9dcdc4c7bfe30")
How it ran

no invocation recorded.

BTNTransformerEnc Not yet run
unverified (harvested, no recorded run)
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("2d11b5a36034202a")
How it ran

no invocation recorded.

DecoderLayer Not yet run
unverified (harvested, no recorded run)
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("137850413fc262a6")
How it ran

no invocation recorded.

OCLR Not yet run
unverified (harvested, no recorded run)
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("09bb9b075e9f690d")
How it ran

no invocation recorded.

TransEncoder Not yet run
unverified (harvested, no recorded run)
Jyxarthur/OCLR_model/models/oclr.py
code served (permissive licence) · get_code("1eceb7d7a75f267f")
How it ran

no invocation recorded.

Also present in other papers' code

Of 21 samples lifted from this paper's linked repositories, 21 carry a function identity and 0 do not. Of the 21 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 0 later) and 20 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
_in_projection_packed
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2205.03776 (2022-05)
SparseTT: Visual Tracking with Sparse Transformers

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

The objective of this paper is a model that is able to discover, track and segment multiple moving objects in a video. We make four contributions: First, we introduce an object-centric segmentation model with a depth-ordered layer representation. This is implemented using a variant of the transformer architecture that ingests optical flow, where each query vector specifies an object and its layer for the entire video. The model can effectively discover multiple moving objects and handle mutual occlusions; Second, we introduce a scalable pipeline for generating multi-object synthetic training data via layer compositions, that is used to train the proposed model, significantly reducing the requirements for labour-intensive annotations, and supporting Sim2Real generalisation; Third, we conduct thorough ablation studies, showing that the model is able to learn object permanence and temporal shape consistency, and is able to predict amodal segmentation masks; Fourth, we evaluate our model, trained only on synthetic data, on standard video segmentation benchmarks, DAVIS, MoCA, SegTrack, FBMS-59, and achieve stateof-the-art performance among existing methods that do not rely on any manual annotations. With test-time adaptation, we observe further performance boosts. 36th Conference on Neural Information Processing Systems (NeurIPS 2022).

For agents

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

get_harvested_code_for_paper("2207.02206")
get_code_for_paper("2207.02206")
have("2207.02206")

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

curl https://syntology.ai/api/ran/2207.02206.json

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

[![Syntology run record](https://syntology.ai/api/ran/2207.02206.svg)](https://syntology.ai/paper/2207.02206)

Connect an agent — have() is free.