SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2108.09964 · ICCV · 2021

Learning Signed Distance Field for Multi-view Surface Reconstruction

Jingyang Zhang, Yao Yao, Long Quan

arXiv · PDF · Open in the Atlas

Code that ran

11 of 12 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (3 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 6 failures of our instrument; 1 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/2108.09964.json · badge: /api/ran/2108.09964.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
jzhangbs/MVSDF marked official in Papers with Code
Papers with Code archive
11 ran · 1 no recorded run
FunctionStatusWhere it lives
ImplicitNetwork Ran
ran on a synthesized input · constructed an object; no result was computed
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("d75c73c5bf9b87d1")
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, ImplicitNetwork: 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 numpy as np

torch.manual_seed(0)

# Helper function needed by ImplicitNetwork
def get_embedder(multires):
    """Positional encoding embedder"""
    freq_bands = torch.linspace(0., multires - 1., steps=multires)
    
    def embed_fn(x):
        # x shape: (batch, 3) -> output shape: (batch, 6*multires)
        encoded = []
        for freq in freq_bands:
            encoded.append(torch.sin((2.0 ** freq) * x))
            encoded.append(torch.cos((2.0 ** freq) * x))
        return torch.cat(encoded, dim=-1)
    
    input_ch = int(2 * multires)
    return embed_fn, input_ch

# Create arguments for ImplicitNetwork
ARGS = {
    'feature_vector_size': 5,      # size of feature vector in output
    'd_in': 3,                      # input dimension (e.g., 3D coordinates)
    'd_out': 2,                     # output dimension (e.g., SDF value)
    'dims': [64, 64, 64],          # hidden layer dimensions
    'geometric_init': True,         # use geometric initialization
    'bias': 1.0,                    # bias initialization value
    'skip_in': (2,),                # skip connections at layer 2
    'weight_norm': True,            # apply weight normalization
    'multires': 0,                  # no positional encoding (set to 0 for simplicity)
}
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 ImplicitNetwork — 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 ImplicitNetwork 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
RayTracing Ran
ran on a synthesized input · constructed an object; no result was computed
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("6f5d947cacc31dd2")
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, RayTracing: 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

torch.manual_seed(0)

# Create a minimal SDF function that the RayTracing module will call
class SimpleSDF(nn.Module):
    def __init__(self):
        super().__init__()
    
    def forward(self, points):
        # points shape: (batch, num_points, 3)
        # Return SDF values
        return torch.norm(points, dim=-1, keepdim=True)

# Create mock for rend_util module since get_sphere_intersection is called
class MockRendUtil:
    @staticmethod
    def get_sphere_intersection(cam_loc, ray_directions, r=1.0):
        n_imgs, n_pix, _ = ray_directions.shape
        cam_loc_expanded = cam_loc.unsqueeze(-1)
        ray_cam_dot = torch.bmm(ray_directions, cam_loc_expanded).squeeze()
        under_sqrt = ray_cam_dot ** 2 - (cam_loc_expanded.norm(2, 1) ** 2 - r ** 2)
        under_sqrt = under_sqrt.reshape(-1)
        mask_intersect = under_sqrt > 0
        sphere_intersections = torch.zeros(n_imgs * n_pix, 2).float()
        sphere_intersections[mask_intersect] = torch.sqrt(under_sqrt[mask_intersect]).unsqueeze(-1) * torch.Tensor([-1, 1]).float()
        sphere_intersections[mask_intersect] -= ray_cam_dot.reshape(-1)[mask_intersect].unsqueeze(-1)
        sphere_intersections = sphere_intersections.reshape(n_imgs, n_pix, 2)
        sphere_intersections = sphere_intersections.clamp_min(0.0)
        mask_intersect = mask_intersect.reshape(n_imgs, n_pix)
        return (sphere_intersections, mask_intersect)

# Mock rend_util globally
import sys
mock_rend_util = MockRendUtil()

class MockModule:
    get_sphere_intersection = staticmethod(MockRendUtil.get_sphere_intersection)

rend_util_mock = MockModule()

# Patch the function to work without CUDA
def get_sphere_intersection_cpu(cam_loc, ray_directions, r=1.0):
    n_imgs, n_pix, _ = ray_directions.shape
    cam_loc_expanded = cam_loc.unsqueeze(-1)
    ray_cam_dot = torch.bmm(ray_directions, cam_loc_expanded).squeeze()
    under_sqrt = ray_cam_dot ** 2 - (cam_loc_expanded.norm(2, 1) ** 2 - r ** 2)
    under_sqrt = under_sqrt.reshape(-1)
    mask_intersect = under_sqrt > 0
    sphere_intersections = torch.zeros(n_imgs * n_pix, 2).float()
    sphere_intersections[mask_intersect] = torch.sqrt(under_sqrt[mask_intersect]).unsqueeze(-1) * torch.Tensor([-1, 1]).float()
    sphere_intersections[mask_intersect] -= ray_cam_dot.reshape(-1)[mask_intersect].unsqueeze(-1)
    sphere_intersections = sphere_intersections.reshape(n_imgs, n_pix, 2)
    sphere_intersections = sphere_intersections.clamp_min(0.0)
    mask_intersect = mask_intersect.reshape(n_imgs, n_pix)
    return (sphere_intersections, mask_intersect)

ARGS = {
    'object_bounding_sphere': 1.0,
    'sdf_threshold': 5e-05,
    'line_search_step': 0.5,
    'line_step_iters': 1,
    'sphere_tracing_iters': 10,
    'n_steps': 100,
    'n_secant_steps': 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 RayTracing — 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 RayTracing 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
RenderingNetwork Ran
ran on a synthesized input · constructed an object; no result was computed
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("fedc19d585ddbef4")
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, RenderingNetwork: 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)

# Mock embedder function for multires_view
def get_embedder(multires):
    """Mock embedder that returns a function and input channel count"""
    def embed_fn(x):
        # Simple mock embedding: duplicate and scale the input
        return torch.cat([x, torch.sin(x), torch.cos(x)], dim=-1)
    
    # multires determines the output dimension
    # For multires=4: we return original + sin + cos = 3x original dim
    input_ch = 3 * x.shape[-1] if hasattr(x, 'shape') else 9
    return embed_fn, 9

# Patch get_embedder into the module scope
import sys
current_module = sys.modules[__name__]
setattr(current_module, 'get_embedder', get_embedder)

# Build tiny test data
batch_size = 2
points_dim = 3
view_dirs_dim = 3
normals_dim = 3
feature_vector_size = 4

# Create sample tensors
points = torch.randn(batch_size, points_dim)
normals = torch.randn(batch_size, normals_dim)
view_dirs = torch.randn(batch_size, view_dirs_dim)
feature_vectors = torch.randn(batch_size, feature_vector_size)

# Define ARGS dict with all constructor parameters
ARGS = {
    'feature_vector_size': feature_vector_size,
    'mode': 'idr',
    'd_in': 3,  # points dimension
    'd_out': 3,  # output dimension (RGB)
    'dims': [32, 32],  # hidden layer dimensions
    'weight_norm': True,
    'multires_view': 0,
}
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 RenderingNetwork — 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 RenderingNetwork 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
SampleNetwork Ran
ran on a synthesized input
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("15fce13e5e90a1a7")
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; 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

class SampleNetwork(nn.Module):
    """
    Represent the intersection (sample) point as differentiable function of the implicit geometry and camera parameters.
    See equation 3 in the paper for more details.
    """

    def forward(self, surface_output, surface_sdf_values, surface_points_grad, surface_dists, surface_cam_loc, surface_ray_dirs):
        surface_ray_dirs_0 = surface_ray_dirs.detach()
        surface_points_dot = torch.bmm(surface_points_grad.view(-1, 1, 3), surface_ray_dirs_0.view(-1, 3, 1)).squeeze(-1)
        surface_dists_theta = surface_dists - (surface_output - surface_sdf_values) / surface_points_dot
        surface_points_theta_c_v = surface_cam_loc + surface_dists_theta * surface_ray_dirs
        return surface_points_theta_c_v


INIT_ARGS = {}

ARGS = {
    'surface_output': torch.randn(2, 1),
    'surface_sdf_values': torch.randn(2, 1),
    'surface_points_grad': torch.randn(2, 3),
    'surface_dists': torch.randn(2, 1),
    'surface_cam_loc': torch.randn(2, 3),
    'surface_ray_dirs': torch.randn(2, 3),
}

model = SampleNetwork(**INIT_ARGS)
output = model(**ARGS)
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 — 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)
get_camera_params Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("94101ffffea9e552")
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. 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.functional as F

torch.manual_seed(0)

# Create UV coordinates: (batch_size=2, num_samples=4, 2)
uv = torch.randint(0, 100, (2, 4, 2)).float()

# Create pose in quaternion format: (batch_size=2, 7)
# First 4 elements are quaternion, last 3 are camera location
pose = torch.randn(2, 7).float()
pose[:, :4] = F.normalize(pose[:, :4], dim=1)  # Normalize quaternion

# Create intrinsics matrix: (batch_size=2, 3, 3)
# Standard camera intrinsics format
intrinsics = torch.zeros(2, 3, 3).float()
intrinsics[:, 0, 0] = 500.0  # fx
intrinsics[:, 1, 1] = 500.0  # fy
intrinsics[:, 0, 2] = 320.0  # cx
intrinsics[:, 1, 2] = 240.0  # cy
intrinsics[:, 2, 2] = 1.0

ARGS = {
    "uv": uv,
    "pose": pose,
    "intrinsics": intrinsics,
}
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
get_pixel_grids Ran
ran, honoured the contract we drafted
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("e3df656c5b4c9cc6")
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 (4, 3, 3, 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

torch.manual_seed(0)

ARGS = {
    "height": 4,
    "width": 3,
    "cuda": 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 Tensor, ndim 4, dtype family float, shape (4, 3, 3, 1) — shape (4, 3, 3, 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
get_sphere_intersection Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("c2d66969848d50fb")
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. 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

torch.manual_seed(0)

# The function expects:
# - cam_loc: tensor of shape (n_imgs, 3) representing camera locations
# - ray_directions: tensor of shape (n_imgs, n_pix, 3) representing ray directions
# - r: scalar radius (default 1.0)

n_imgs = 2
n_pix = 4

# cam_loc should be (n_imgs, 3)
cam_loc = torch.randn(n_imgs, 3)

# ray_directions should be (n_imgs, n_pix, 3)
ray_directions = torch.randn(n_imgs, n_pix, 3)

# r is a scalar radius
r = 1.0

ARGS = {
    "cam_loc": cam_loc,
    "ray_directions": ray_directions,
    "r": r
}
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
idx_cam2world Ran
ran; our contract draft was wrong, not the code
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("09bde7e60b873349")
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 (2, 3, 5, 4, 1) 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)

# idx_cam_homo: nhw41 tensor (n=batch, h=height, w=width, 4=homogeneous coords, 1=feature dim)
# The function indexes into it with [..., -1:, :] and performs matrix operations
# Shape should be (n, h, w, 4, 1)
idx_cam_homo = torch.randn(2, 3, 5, 4, 1)

# cam: should support indexing cam[:, 0:1, ...] and .inverse()
# The result cam[:, 0:1, ...] needs to be a matrix that can be inverted
# cam[:, 0:1, ...] produces shape (batch, 1, ...) and needs to support .inverse()
# This suggests cam is a batch of matrices. Given the unsqueeze(1), cam should be (batch, num_matrices, 4, 4)
# where [:, 0:1, ...] selects the first matrix for each batch element
cam = torch.randn(2, 3, 4, 4)
# Make sure the matrices are invertible by using orthogonal matrices
cam = torch.linalg.qr(torch.randn(2, 3, 4, 4))[0]

ARGS = {
    'idx_cam_homo': idx_cam_homo,
    'cam': cam,
}
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 5, dtype family float, shape (2, 3, 5, 4, 1) — shape (2, 3, 5, 4, 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
idx_img2cam Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("1708699fc850cd66")
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 (2, 4, 5, 4, 1) 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)

# idx_img_homo: nhw31 - batch_size=2, height=4, width=5, 3 homogeneous coords, 1 extra dim
idx_img_homo = torch.randn(2, 4, 5, 3, 1)

# depth: n1hw - batch_size=2, 1 channel, height=4, width=5
depth = torch.randn(2, 1, 4, 5)

# cam: camera matrix with shape that allows indexing cam[:, 1:2, :3, :3]
# cam should be at least (batch_size, >=2, >=3, >=3)
# Using shape (2, 3, 4, 4) to have batch, and sufficient spatial dims
cam = torch.randn(2, 3, 4, 4)

ARGS = {
    "idx_img_homo": idx_img_homo,
    "depth": depth,
    "cam": cam,
}
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 5, dtype family float, shape (2, 4, 5, 4, 1) — shape (2, 4, 5, 4, 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
lift Ran
ran; our contract draft was wrong, not the code
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("24775db8fc54253e")
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; 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 torch

torch.manual_seed(0)

# x: pixel x coordinates, shape (batch, num_points)
x = torch.rand(2, 4)

# y: pixel y coordinates, shape (batch, num_points)
y = torch.rand(2, 4)

# z: depth values, shape (batch, num_points)
z = torch.rand(2, 4) + 0.5  # avoid near-zero depths

# intrinsics: camera intrinsic matrix, shape (batch, 3, 3)
# Standard form: [[fx, sk, cx], [0, fy, cy], [0, 0, 1]]
intrinsics = torch.zeros(2, 3, 3)
intrinsics[:, 0, 0] = torch.tensor([500.0, 480.0])  # fx
intrinsics[:, 1, 1] = torch.tensor([500.0, 480.0])  # fy
intrinsics[:, 0, 2] = torch.tensor([320.0, 320.0])  # cx
intrinsics[:, 1, 2] = torch.tensor([240.0, 240.0])  # cy
intrinsics[:, 0, 1] = torch.tensor([0.0, 0.0])      # skew
intrinsics[:, 2, 2] = 1.0

ARGS = {
    "x": x,
    "y": y,
    "z": z,
    "intrinsics": intrinsics,
}
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 — 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
quat_to_rot Ran
ran; our contract draft was wrong, not the code
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("7e56dae01ba2f402")
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; 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 torch
import torch.nn.functional as F

torch.manual_seed(0)

# Create a batch of quaternions
# Shape: (batch_size=2, quat_dims=4)
# Quaternions are typically [w, x, y, z] or [x, y, z, w]
q = torch.randn(2, 4)

ARGS = {
    "q": q
}
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 — 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
IDRNetwork Not yet run
unverified (harvested, no recorded run)
jzhangbs/MVSDF/code/model/implicit_differentiable_renderer.py
code served (permissive licence) · get_code("5f8bf3bd453e7dfe")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 5 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 3 later) and 7 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
lift
brick of this paper's assembly: helper
2 other papers
2 later · 2 only as an assembly brick
2207.09686 (2022-07)
Object-Compositional Neural Implicit Surfaces
2303.08605 (2023-03)
RICO: Regularizing the Unobservable for Indoor Compositional Reconstruction
quat_to_rot
brick of this paper's assembly: helper
2 other papers
2 later · 1 only as an assembly brick
2207.09686 (2022-07)
Object-Compositional Neural Implicit Surfaces
2303.08605 (2023-03)
RICO: Regularizing the Unobservable for Indoor Compositional Reconstruction
get_sphere_intersection
brick of this paper's assembly: helper
1 other paper
1 later
2308.11025 (2023-08)
Coordinate Quantized Neural Implicit Representations for Multi-view Reconstruction
idx_cam2world
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2008.07928 (2020-08)
Visibility-aware Multi-view Stereo Network
idx_img2cam
brick of this paper's assembly: helper
1 other paper
1 dated earlier
2008.07928 (2020-08)
Visibility-aware Multi-view Stereo Network

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

Recent works on implicit neural representations have shown promising results for multi-view surface reconstruction. However, most approaches are limited to relatively simple geometries and usually require clean object masks for reconstructing complex and concave objects. In this work, we introduce a novel neural surface reconstruction framework that leverages the knowledge of stereo matching and feature consistency to optimize the implicit surface representation. More specifically, we apply a signed distance field (SDF) and a surface light field to represent the scene geometry and appearance respectively. The SDF is directly supervised by geometry from stereo matching, and is refined by optimizing the multi-view feature consistency and the fidelity of rendered images. Our method is able to improve the robustness of geometry estimation and support reconstruction of complex scene topologies. Extensive experiments have been conducted on DTU, EPFL and Tanks and Temples datasets. Compared to previous state-of-theart methods, our method achieves better mesh reconstruction in wide open scenes without masks as input.

For agents

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

get_harvested_code_for_paper("2108.09964")
get_code_for_paper("2108.09964")
have("2108.09964")

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

curl https://syntology.ai/api/ran/2108.09964.json

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

[![Syntology run record](https://syntology.ai/api/ran/2108.09964.svg)](https://syntology.ai/paper/2108.09964)

Connect an agent — have() is free.