SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.10087 · 2023

Neural Refinement for Absolute Pose Regression with Feature Synthesis

arXiv · PDF · Open in the Atlas

Code that ran

17 of 17 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 2 honoured the contract we drafted, 1 violated it and 9 have no contract verdict; 5 failures of our instrument; 0 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/2303.10087.json · badge: /api/ran/2303.10087.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
ActiveVisionLab/NeFeS canonical (classifier label)
classifier basis not recorded on the link
17 ran · 0 no recorded run
FunctionStatusWhere it lives
camera_frustum_initializer Ran
ran; our contract draft was wrong, not the code
ActiveVisionLab/NeFeS/dataset_loaders/load_7Scenes.py
code served (permissive licence) · get_code("c9625d568955ffac")
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 numpy as np

class Args:
    def __init__(self, near_far):
        self.near_far = near_far

ARGS = {
    "args": Args(near_far=[0.1, 5.0])
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — 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
compute_pose_error_SE3 Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/script/dm/pose_model.py
code served (permissive licence) · get_code("357b4ed9c9213d8f")
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 1, 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
import math
import cv2

torch.manual_seed(0)

# Create two sample SE(3) pose matrices
# SE(3) matrices are 4x4 (or 3x4) with rotation in top-left 3x3 and translation in top-right 3x1

# First pose matrix (4x4)
pose = torch.eye(4)
pose[0:3, 0:3] = torch.tensor([
    [0.9, -0.436, 0.0],
    [0.436, 0.9, 0.0],
    [0.0, 0.0, 1.0]
])
pose[0:3, 3] = torch.tensor([1.0, 2.0, 3.0])

# Second pose matrix (4x4) - slightly different
predict_pose = torch.eye(4)
predict_pose[0:3, 0:3] = torch.tensor([
    [0.891, -0.454, 0.0],
    [0.454, 0.891, 0.0],
    [0.0, 0.0, 1.0]
])
predict_pose[0:3, 3] = torch.tensor([1.1, 2.1, 3.2])

ARGS = {
    "pose": pose,
    "predict_pose": predict_pose
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 1, 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
disable_model_grad Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/script/dm/direct_pose_model.py
code served (permissive licence) · get_code("5ebe28260a0e2f2e")
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 SimpleNeRFModel. 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 SimpleNeRFModel(nn.Module):
    """Minimal NeRF-like model with weight and bias parameters"""
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
        self.bn = nn.BatchNorm1d(2)
    
    def forward(self, x):
        x = self.linear1(x)
        x = self.linear2(x)
        x = self.bn(x)
        return x

model = SimpleNeRFModel()

ARGS = {
    'model': model
}
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 SimpleNeRFModel — a SimpleNeRFModel output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SimpleNeRFModel 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
filter_hook Ran
ran; our contract draft was wrong, not the code
ActiveVisionLab/NeFeS/script/dm/pose_model.py
code served (permissive licence) · get_code("c7bea2e25b534df3")
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 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

torch.manual_seed(0)

# The function filter_hook is a backward hook that processes input gradients
# m: the module (not directly used in the function body)
# g_in: tuple of input gradients to the module
# g_out: tuple of output gradients from the module (not used in the function body)

# Create a simple module
m = nn.Linear(4, 3)

# Create gradient tensors for g_in
# The function iterates over g_in, clones each tensor, replaces NaN values with 0
# We need to create tensors that could contain NaNs
g_tensor_1 = torch.randn(2, 4)
g_tensor_2 = torch.randn(2, 4)

# Introduce some NaN values to make the filtering meaningful
g_tensor_1[0, 0] = float('nan')
g_tensor_2[1, 2] = float('nan')

# g_in should be a tuple of gradients
g_in = (g_tensor_1, g_tensor_2)

# g_out is not used in the function body, but it's a parameter
# Typically it's a tuple of output gradients
g_out = (torch.randn(2, 3),)

ARGS = {
    "m": m,
    "g_in": g_in,
    "g_out": g_out
}
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 3, 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
generate_sampling_frustum Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/dataset_loaders/load_Cambridge.py
code served (permissive licence) · get_code("1fc1ac3d7fc531a9")
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 ndarray, ndim 2, dtype family float, shape (367, 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 numpy as np

# Generate setup code for is_inside_frustum and generate_sampling_frustum

# Analyze the function:
# - step: a number (sampling step size)
# - depth: a number (maximum depth)
# - K: a 3x3 camera intrinsic matrix (used with np.dot)
# - f: focal length (a number)
# - cx, cy: camera principal point (numbers)
# - x_res, y_res: image resolution in pixels (numbers)

# Create a simple camera intrinsic matrix K
# Standard form:
# K = [[f, 0, cx],
#      [0, f, cy],
#      [0, 0, 1]]
K = np.array([
    [500.0, 0.0, 320.0],
    [0.0, 500.0, 240.0],
    [0.0, 0.0, 1.0]
], dtype=np.float32)

# Parameters
step = 0.5  # sampling step size in meters
depth = 5.0  # maximum depth in meters
f = 500.0  # focal length in pixels
cx = 320.0  # principal point x
cy = 240.0  # principal point y
x_res = 640  # image width in pixels
y_res = 480  # image height in pixels

ARGS = {
    'step': step,
    'depth': depth,
    'K': K,
    'f': f,
    'cx': cx,
    'cy': cy,
    'x_res': x_res,
    'y_res': y_res
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (367, 3) — shape (367, 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
inference_pose_regression Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/script/dm/direct_pose_model.py
code served (permissive licence) · get_code("70a610a299c24c6d")
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 (2, 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

torch.manual_seed(0)

# Create a minimal PoseNet model that outputs 12 values (3x4 flattened)
class MinimalPoseNet(nn.Module):
    def __init__(self):
        super().__init__()
        self.fc = nn.Linear(12, 12)  # Input: 2 channels × 2 height × 3 width = 12
    
    def forward(self, x):
        # Flatten input and output 12 values
        batch_size = x.shape[0]
        x = x.view(batch_size, -1)
        return self.fc(x)

# Create a minimal args object
class Args:
    def __init__(self):
        self.batch_size = 2
        self.svd_reg = True

# Set up the components
args = Args()
device = torch.device('cpu')
model = MinimalPoseNet()
model.eval()

# Create input data: (batch_size=2, channels=2, H=2, W=3)
data = torch.randn(2, 2, 2, 3)

ARGS = {
    'args': args,
    'data': data,
    'device': device,
    'model': model
}
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 3, dtype family float, shape (2, 3, 4) — shape (2, 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
initK Ran
ran, honoured the contract we drafted
ActiveVisionLab/NeFeS/dataset_loaders/load_7Scenes.py
code served (permissive licence) · get_code("13be307ef0c24c0e")
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 ndarray, ndim 2, dtype family float, shape (3, 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 numpy as np

ARGS = {
    "f": 500.0,
    "cx": 320.0,
    "cy": 240.0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (3, 3) — shape (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
is_inside_frustum Ran
ran, violated the contract we drafted
ActiveVisionLab/NeFeS/dataset_loaders/load_7Scenes.py
code served (permissive licence) · get_code("f869aad3a90d4156")
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 1, dtype family float, shape (2,) 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)

# Analyzing the function:
# - p: appears to be indexable with [0] and [1], likely a tensor or array
# - x_res: a scalar representing x resolution (compared with p[0])
# - y_res: a scalar representing y resolution (compared with p[1])
# The function checks if point p is within bounds (0, x_res) and (0, y_res)

# Create a point tensor with 2 coordinates
p = torch.tensor([[2.5, 3.5], [0.5, 4.5]], dtype=torch.float32)  # shape (2, 2)

# Resolution values
x_res = 10
y_res = 8

ARGS = {
    "p": p,
    "x_res": x_res,
    "y_res": y_res
}
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 1, dtype family float, shape (2,) — shape (2,) 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
load_image Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/dataset_loaders/cambridge_scenes.py
code served (permissive licence) · get_code("9b5a8fceac6859e6")
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 Image. 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 tempfile
from PIL import Image
import os
from torchvision.datasets.folder import default_loader

# Create a temporary image file for testing
temp_dir = tempfile.mkdtemp()
temp_image_path = os.path.join(temp_dir, "test_image.jpg")

# Create a simple test image using PIL
test_image = Image.new('RGB', (10, 10), color='red')
test_image.save(temp_image_path)

# Define ARGS dict with all parameters
ARGS = {
    'filename': temp_image_path,
    'loader': default_loader
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, torch, torchvision — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Image — a Image output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Image output the recorded ndim 3 and dtype family int 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
perturb_rotation Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/dataset_loaders/load_7Scenes_colmap.py
code served (permissive licence) · get_code("e0dc10227c25a9e9")
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 ndarray, ndim 2, dtype family float, shape (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 numpy as np

# c2w should be a 3x4 matrix (camera-to-world transformation)
c2w = np.array([
    [1.0, 0.0, 0.0, 0.0],
    [0.0, 1.0, 0.0, 0.0],
    [0.0, 0.0, 1.0, 5.0]
], dtype=float)

# theta: rotation angle in degrees around y-axis
theta = 45.0

# phi: rotation angle in degrees around x-axis
phi = 30.0

# psi: rotation angle in degrees around z-axis (optional, defaults to 0)
psi = 15.0

ARGS = {
    "c2w": c2w,
    "theta": theta,
    "phi": phi,
    "psi": psi
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (3, 4) — shape (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
preprocess_data Ran
ran; our contract draft was wrong, not the code
ActiveVisionLab/NeFeS/script/dm/pose_model.py
code served (permissive licence) · get_code("1d4f466dd0b12111")
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, 4, 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

torch.manual_seed(0)

# Create a small input tensor: batch size 2, 3 channels, 4x4 spatial dimensions
inputs = torch.randn(2, 3, 4, 4)

# Set device to CPU
device = torch.device('cpu')

ARGS = {
    'inputs': inputs,
    'device': device
}
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, 4, 4) — shape (2, 3, 4, 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
process_poses_rotmat Ran
ran, honoured the contract we drafted
ActiveVisionLab/NeFeS/dataset_loaders/seven_scenes.py
code served (permissive licence) · get_code("ecb54915a44f88b3")
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 (2, 12) 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)

# Construct arguments for process_poses_rotmat
# poses_in: N x 12 tensor of raw poses
poses_in = torch.randn(2, 12)

# mean_t: mean translation vector (typically 3D)
mean_t = torch.randn(3)

# std_t: standard deviation for translation (typically 3D)
std_t = torch.randn(3).abs() + 0.1  # ensure positive

# align_R: alignment rotation matrix (typically 3x3)
align_R = torch.eye(3)

# align_t: alignment translation vector (typically 3D)
align_t = torch.randn(3)

# align_s: alignment scale (scalar)
align_s = torch.tensor(1.0)

ARGS = {
    "poses_in": poses_in,
    "mean_t": mean_t,
    "std_t": std_t,
    "align_R": align_R,
    "align_t": align_t,
    "align_s": align_s,
}
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 2, dtype family float, shape (2, 12) — shape (2, 12) 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
process_poses_rotmat Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/dataset_loaders/cambridge_scenes.py
code served (permissive licence) · get_code("aba0c9bab294b165")
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 ndarray, ndim 2, dtype family float, shape (2, 12) 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 numpy as np

# Create small sample data
batch_size = 2

# poses_in: N x 7 (position 3D + quaternion 4D)
poses_in = np.random.randn(batch_size, 7).astype(np.float32)

# rot_mat: N x 3 x 3 (rotation matrices)
rot_mat = np.random.randn(batch_size, 3, 3).astype(np.float32)

ARGS = {
    "poses_in": poses_in,
    "rot_mat": rot_mat,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 12) — shape (2, 12) 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
qlog Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ActiveVisionLab/NeFeS/dataset_loaders/cambridge_scenes.py
code served (permissive licence) · get_code("667f7d75349dad79")
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 ndarray, ndim 1, dtype family float, shape (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 numpy as np

# The function qlog takes a quaternion q of shape (4,) and applies a logarithm map
# to return a vector of shape (3,)
# 
# Looking at the function body:
# - q is indexed with q[0], q[1:], and np.linalg.norm(q[1:])
# - It's a 1D numpy array of length 4

# Create a valid quaternion input
# A quaternion is typically [w, x, y, z] where w is the scalar part
# We'll create one with a non-zero imaginary part to avoid the zero-check path

q_value = np.array([0.7071, 0.7071, 0.0, 0.0], dtype=np.float64)

ARGS = {
    "q": q_value
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 1, dtype family float, shape (3,) — shape (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
svd_reg Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/script/dm/direct_pose_model.py
code served (permissive licence) · get_code("862a1ba2f5abb651")
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 (2, 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

torch.manual_seed(0)

# The function expects:
# - pose: a tensor of shape [B, 3, 4] representing batch of 3x4 pose matrices
# The function extracts the 3x3 rotation part, performs SVD, and regularizes it

# Create a batch of 2 pose matrices, each 3x4
# Shape: [batch_size=2, 3, 4]
pose = torch.randn(2, 3, 4)

ARGS = {
    "pose": pose
}
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 3, dtype family float, shape (2, 3, 4) — shape (2, 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
trunc_softplus Ran
ran on a synthesized input
ActiveVisionLab/NeFeS/script/models/activation.py
code served (permissive licence) · get_code("10dd26fa704c4ec8")
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 (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

torch.manual_seed(0)

# The function takes a single parameter 'x' which should be a tensor
# It clamps the tensor and applies softplus, so x needs to be a tensor
x = torch.randn(2, 4)

ARGS = {
    "x": x
}
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 (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: none by the harness; any seed is in the setup block
viewmatrix Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
ActiveVisionLab/NeFeS/dataset_loaders/load_7Scenes_colmap.py
code served (permissive licence) · get_code("562861bc16dfb5c0")
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 ndarray, ndim 2, dtype family float, shape (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 numpy as np

# Analyze the function usage:
# - z: passed to normalize(), then used in np.cross() and np.stack()
# - up: used in np.cross() with z
# - pos: used in np.stack() as the last element

# All parameters are 1D vectors that go through normalize() or direct use
# np.cross() expects 3D vectors, and the result is stacked to form a matrix

# Create 3D vectors for cross product operations
z = np.array([1.0, 0.0, 0.0])
up = np.array([0.0, 1.0, 0.0])
pos = np.array([0.0, 0.0, 5.0])

ARGS = {
    "z": z,
    "up": up,
    "pos": pos,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (3, 4) — shape (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

Also present in other papers' code

Of 17 samples lifted from this paper's linked repositories, 17 carry a function identity and 0 do not. Of the 17 distinct functions with an identity, 12 are also present in other papers whose code we hold (12 with at least one paper dated earlier, 0 the same month, 1 later) and 5 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
viewmatrix 17 other papers
10 dated earlier, 6 later, 1 same year, order not determinable
2003.08934 (2020-03)
NeRF: Representing Scenes as Neural Radiance Fields for View Synthesis
2007.02442 (2020-07)
GRAF: Generative Radiance Fields for 3D-Aware Image Synthesis
2104.04073 (2021-04)
Direct-PoseNet: Absolute Pose Regression with Photometric Consistency
and 14 more not named here
camera_frustum_initializer 2 other papers
2 dated earlier
2104.04073 (2021-04)
Direct-PoseNet: Absolute Pose Regression with Photometric Consistency
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
filter_hook 2 other papers
2 dated earlier
2104.04073 (2021-04)
Direct-PoseNet: Absolute Pose Regression with Photometric Consistency
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
initK 2 other papers
2 dated earlier
2104.04073 (2021-04)
Direct-PoseNet: Absolute Pose Regression with Photometric Consistency
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
is_inside_frustum 2 other papers
2 dated earlier
2104.04073 (2021-04)
Direct-PoseNet: Absolute Pose Regression with Photometric Consistency
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
preprocess_data 2 other papers
2 dated earlier
2104.04073 (2021-04)
Direct-PoseNet: Absolute Pose Regression with Photometric Consistency
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
process_poses_rotmat 2 other papers
2 dated earlier
2104.04073 (2021-04)
Direct-PoseNet: Absolute Pose Regression with Photometric Consistency
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
qlog 2 other papers
2 dated earlier
2104.04073 (2021-04)
Direct-PoseNet: Absolute Pose Regression with Photometric Consistency
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
disable_model_grad 1 other paper
1 dated earlier
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
generate_sampling_frustum 1 other paper
1 dated earlier
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
load_image 1 other paper
1 dated earlier
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching
process_poses_rotmat 1 other paper
1 dated earlier
2204.00559 (2022-04)
DFNet: Enhance Absolute Pose Regression with Direct Feature Matching

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

Absolute Pose Regression (APR) methods use deep neural networks to directly regress camera poses from RGB images. However, the predominant APR architectures only rely on 2D operations during inference, resulting in limited accuracy of pose estimation due to the lack of 3D geometry constraints or priors. In this work, we propose a test-time refinement pipeline that leverages implicit geometric constraints using a robust feature field to enhance the ability of APR methods to use 3D information during inference. We also introduce a novel Neural Feature Synthesizer (NeFeS) model, which encodes 3D geometric features during training and directly renders dense novel view features at test time to refine APR methods. To enhance the robustness of our model, we introduce a feature fusion module and a progressive training strategy. Our proposed method achieves state-of-the-art single-image APR accuracy on indoor and outdoor datasets.

For agents

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

get_harvested_code_for_paper("2303.10087")
get_code_for_paper("2303.10087")
have("2303.10087")

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

curl https://syntology.ai/api/ran/2303.10087.json

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

[![Syntology run record](https://syntology.ai/api/ran/2303.10087.svg)](https://syntology.ai/paper/2303.10087)

Connect an agent — have() is free.