SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2409.03634 · ECCV · 2024

Surface-Centric Modeling for High-Fidelity Generalizable Neural Surface Reconstruction

Jianbo Jiao, Xiaodong Gu, Ronggang Wang, Rui Peng, Shihe Shen, Kaiqiang Xiong, Huachen Gao

arXiv · PDF · Open in the Atlas

Code that ran

5 of 8 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; 1 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 4 failures of our instrument; 3 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/2409.03634.json · badge: /api/ran/2409.03634.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
prstrive/SuRF canonical (classifier label)
classifier basis: census
5 ran · 3 no recorded run
FunctionStatusWhere it lives
compute_consistency_loss Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
prstrive/SuRF/models/losses/consistency_loss.py
code served (permissive licence) · get_code("635d49f96eed67d1")
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 0, dtype family float, shape () 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)

# Small tensor dimensions for depth maps
height, width = 4, 5

# Reference depth map (4x5)
ref_depth = torch.rand(height, width, dtype=torch.float32) + 0.5

# Source depth map (4x5)
src_depth = torch.rand(height, width, dtype=torch.float32) + 0.5

# Intrinsic matrices (3x3 each)
# Using distinct values to differentiate from other matrices
ref_intr = torch.tensor([
    [500.0, 0.0, 2.5],
    [0.0, 500.0, 2.0],
    [0.0, 0.0, 1.0]
], dtype=torch.float32)

src_intr = torch.tensor([
    [480.0, 0.0, 2.4],
    [0.0, 480.0, 1.9],
    [0.0, 0.0, 1.0]
], dtype=torch.float32)

# Intrinsics list
intrs = [ref_intr, src_intr]

# Camera-to-world matrices (4x4 each)
# Reference camera at origin
ref_c2w = torch.tensor([
    [1.0, 0.0, 0.0, 0.0],
    [0.0, 1.0, 0.0, 0.0],
    [0.0, 0.0, 1.0, 0.0],
    [0.0, 0.0, 0.0, 1.0]
], dtype=torch.float32)

# Source camera slightly offset
src_c2w = torch.tensor([
    [0.99, -0.1, 0.0, 0.5],
    [0.1, 0.99, 0.0, 0.3],
    [0.0, 0.0, 1.0, 1.0],
    [0.0, 0.0, 0.0, 1.0]
], dtype=torch.float32)

# C2W list
c2ws = [ref_c2w, src_c2w]

# Source image index
src_idx = 1

# Reference depth mask (all ones, 4x5)
mask_ref = torch.ones(height, width, dtype=torch.float32)

# Source depth mask (all ones, 4x5)
mask_src = torch.ones(height, width, dtype=torch.float32)

ARGS = {
    "ref_depth": ref_depth,
    "src_depth": src_depth,
    "intrs": intrs,
    "c2ws": c2ws,
    "src_idx": src_idx,
    "mask_ref": mask_ref,
    "mask_src": mask_src,
}
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 0, dtype family float, shape () — shape () 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
gen_rays_from_single_image Ran
ran; our contract draft was wrong, not the code
prstrive/SuRF/evaluation/clean_mesh.py
code served (permissive licence) · get_code("baaad4f7794fa092")
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 dict. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

H = 4
W = 6
image = torch.randn(3, H, W)
intrinsic = torch.tensor([
    [500.0, 0.0, W/2],
    [0.0, 500.0, H/2],
    [0.0, 0.0, 1.0]
])
c2w = torch.eye(4)
depth = torch.randn(H, W, 1)
mask = torch.ones(H, W, 1)

ARGS = {
    'H': H,
    'W': W,
    'image': image,
    'intrinsic': intrinsic,
    'c2w': c2w,
    'depth': depth,
    'mask': mask
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type dict — a dict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a dict output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
load_K_Rt_from_P Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
prstrive/SuRF/datasets/dtu_finetune.py
code served (permissive licence) · get_code("c6324b632487a725")
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 cv2
import numpy as np

# Create a valid 3x4 projection matrix P
# This is a typical camera projection matrix format
P = np.array([
    [500.0, 0.0, 320.0, 0.0],
    [0.0, 500.0, 240.0, 0.0],
    [0.0, 0.0, 1.0, 0.0]
], dtype=np.float32)

ARGS = {
    "filename": None,
    "P": P
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — 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
load_K_Rt_from_P Ran
ran, honoured the contract we drafted
prstrive/SuRF/evaluation/clean_mesh.py
code served (permissive licence) · get_code("7a8bf46877ee62d3")
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 numpy as np
import cv2 as cv
import tempfile
import os

# Create a temporary file with projection matrix data
# A valid P matrix is 3x4
temp_dir = tempfile.gettempdir()
temp_file = os.path.join(temp_dir, "test_P_matrix.txt")

# Create a simple valid projection matrix (3x4)
# Format: 3 lines with 4 space-separated float values each
P_matrix = np.array([
    [1.0, 0.0, 0.0, 0.0],
    [0.0, 1.0, 0.0, 0.0],
    [0.0, 0.0, 1.0, 1.0]
], dtype=np.float32)

with open(temp_file, 'w') as f:
    for row in P_matrix:
        f.write(' '.join(str(x) for x in row) + '\n')

ARGS = {
    "filename": temp_file,
    "P": None
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
cv2, numpy — 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
read_pfm Ran
ran; our contract draft was wrong, not the code
prstrive/SuRF/datasets/dtu_finetune.py
code served (permissive licence) · get_code("18ed79bdb5f6cf13")
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 tempfile
import numpy as np
import os

# Create a temporary PFM file for testing
def create_test_pfm(filename, width=4, height=3, color=True, scale=1.0):
    """Create a minimal valid PFM file for testing"""
    with open(filename, 'wb') as f:
        # Write header
        header = 'PF' if color else 'Pf'
        f.write(header.encode('utf-8') + b'\n')
        
        # Write dimensions
        f.write(f'{width} {height}\n'.encode('utf-8'))
        
        # Write scale
        f.write(f'{scale}\n'.encode('utf-8'))
        
        # Write data
        if color:
            data = np.random.rand(height, width, 3).astype(np.float32)
        else:
            data = np.random.rand(height, width).astype(np.float32)
        
        # Flip data vertically to match PFM format
        data = np.flipud(data)
        
        # Write as big-endian float
        data.astype('>f').tofile(f)

# Create temporary PFM file
temp_dir = tempfile.gettempdir()
test_pfm_path = os.path.join(temp_dir, 'test_image.pfm')
create_test_pfm(test_pfm_path, width=4, height=3, color=True, scale=1.0)

ARGS = {
    'filename': test_pfm_path
}
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
load_and_decompress Not yet run
unverified (harvested, no recorded run)
prstrive/SuRF/datasets/dtu.py
code served (permissive licence) · get_code("c685f55753cd500b")
How it ran

no invocation recorded.

read_cam_file Not yet run
unverified (harvested, no recorded run)
prstrive/SuRF/evaluation/clean_mesh.py
code served (permissive licence) · get_code("0b9fa356fe48ceef")
How it ran

no invocation recorded.

sample_single_tri Not yet run
unverified (harvested, no recorded run)
prstrive/SuRF/evaluation/dtu_eval.py
code served (permissive licence) · get_code("7ceb5dd3dcfedef1")
How it ran

no invocation recorded.

Also present in other papers' code

Of 8 samples lifted from this paper's linked repositories, 8 carry a function identity and 0 do not. Of the 8 distinct functions with an identity, 6 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 0 the same month, 3 later) and 2 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
load_K_Rt_from_P 14 other papers
13 dated earlier, 1 later
2106.10689 (2021-06)
NeuS: Learning Neural Implicit Surfaces by Volume Rendering for Multi-view Reconstruction
2201.05989 (2022-01)
Instant neural graphics primitives with a multiresolution hash encoding
2205.15848 (2022-05)
Geo-Neus: Geometry-Consistent Neural Implicit Surfaces Learning for Multi-view Reconstruction
and 11 more not named here
read_pfm 12 other papers
11 dated earlier, 1 later
1912.06378 (2019-12)
Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching
2008.02268 (2020-08)
NeRF in the Wild: Neural Radiance Fields for Unconstrained Photo Collections
2108.03824 (2021-08)
AA-RMVSNet: Adaptive Aggregation Recurrent Multi-view Stereo Network
and 9 more not named here
sample_single_tri 7 other papers
6 dated earlier, 1 later
2205.15848 (2022-05)
Geo-Neus: Geometry-Consistent Neural Implicit Surfaces Learning for Multi-view Reconstruction
2208.10925 (2022-08)
Vox-Surf: Voxel-based Implicit Surface Representation
Ren_VolRecon_Volume_Rendering_of_Signed_Ray_Distance_Functions_for_Generalizable_CVPR_2023_paper (2023 (year only))
VolRecon: Volume Rendering of Signed Ray Distance Functions for Generalizable Multi-View Reconstruction
and 4 more not named here
load_K_Rt_from_P 5 other papers
4 dated earlier, 1 same year, order not determinable
2106.12052 (2021-06)
Volume Rendering of Neural Implicit Surfaces
2312.06657 (2023-12)
Learning Naturally Aggregated Appearance for Efficient 3D Editing
2312.12726 (2023-12)
Reducing Shape-Radiance Ambiguity in Radiance Fields with a Closed-Form Color Estimation Method
and 2 more not named here
gen_rays_from_single_image 4 other papers
4 dated earlier
Ren_VolRecon_Volume_Rendering_of_Signed_Ray_Distance_Functions_for_Generalizable_CVPR_2023_paper (2023 (year only))
VolRecon: Volume Rendering of Signed Ray Distance Functions for Generalizable Multi-View Reconstruction
2403.05086 (2024-03)
UFORecon: Generalizable Sparse-View Surface Reconstruction from Arbitrary and Unfavorable Sets
2403.11131 (2024-03)
Omni-Recon: Harnessing Image-based Rendering for General-Purpose Neural Radiance Fields
and 1 more not named here
read_cam_file 3 other papers
3 dated earlier
Ren_VolRecon_Volume_Rendering_of_Signed_Ray_Distance_Functions_for_Generalizable_CVPR_2023_paper (2023 (year only))
VolRecon: Volume Rendering of Signed Ray Distance Functions for Generalizable Multi-View Reconstruction
2403.11131 (2024-03)
Omni-Recon: Harnessing Image-based Rendering for General-Purpose Neural Radiance Fields
2406.02495 (2024-06)
GenS: Generalizable Neural Surface Reconstruction from Multi-View Images

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

Reconstructing the high-fidelity surface from multi-view images, especially sparse images, is a critical and practical task that has attracted widespread attention in recent years. However, existing methods are impeded by the memory constraint or the requirement of groundtruth depths and cannot recover satisfactory geometric details. To this end, we propose SuRF, a new Surface-centric framework that incorporates a new Region sparsification based on a matching Field, achieving good trade-offs between performance, efficiency and scalability. To our knowledge, this is the first unsupervised method achieving end-to-end sparsification powered by the introduced matching field, which leverages the weight distribution to efficiently locate the boundary regions containing surface. Instead of predicting an SDF value for each voxel, we present a new region sparsification approach to sparse the volume by judging whether the voxel is inside the surface region. In this way, our model can exploit higher frequency features around the surface with less memory and computational consumption. Extensive experiments on multiple benchmarks containing complex large-scale scenes show that our reconstructions exhibit high-quality details and achieve new stateof-the-art performance, i.e., 46% improvements with 80% less memory consumption. Code is available at https://github.com/prstrive/SuRF.

For agents

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

get_harvested_code_for_paper("2409.03634")
get_code_for_paper("2409.03634")
have("2409.03634")

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

curl https://syntology.ai/api/ran/2409.03634.json

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

[![Syntology run record](https://syntology.ai/api/ran/2409.03634.svg)](https://syntology.ai/paper/2409.03634)

Connect an agent — have() is free.