SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2104.00681 · CVPR · 2021

NeuralRecon: Real-Time Coherent 3D Reconstruction from Monocular Video

Hujun Bao, Jiaming Sun, Xiaowei Zhou, Linghao Chen, Yiming Xie

arXiv · PDF · Open in the Atlas

Code that ran

7 of 7 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; 0 honoured the contract we drafted, 0 violated it and 0 have no contract verdict; 7 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/2104.00681.json · badge: /api/ran/2104.00681.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
zju3dv/NeuralRecon canonical (classifier label)
classifier basis: census
7 ran · 0 no recorded run
FunctionStatusWhere it lives
back_project Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zju3dv/NeuralRecon/ops/back_project.py
code served (permissive licence) · get_code("2bd47d1b290f0f45")
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)

# Parameters
n_views = 2
batch_size = 2
num_voxels = 8
c = 16  # feature channels
h = 32  # image height
w = 40  # image width

# coords: (num_voxels, 4) with batch indices and (x, y, z) coordinates
coords = torch.zeros(num_voxels, 4, dtype=torch.long)
coords[:4, 0] = 0  # batch 0
coords[4:, 0] = 1  # batch 1
coords[:, 1] = torch.randint(0, 4, (num_voxels,))  # x
coords[:, 2] = torch.randint(0, 4, (num_voxels,))  # y
coords[:, 3] = torch.randint(0, 4, (num_voxels,))  # z

# origin: (batch_size, 3) - origin of voxel volume for each batch
origin = torch.tensor([
    [-1.0, -1.0, -1.0],
    [0.0, 0.0, 0.0]
], dtype=torch.float32)

# voxel_size: scalar
voxel_size = 0.1

# feats: (n_views, batch_size, c, h, w)
feats = torch.randn(n_views, batch_size, c, h, w, dtype=torch.float32)

# KRcam: (n_views, batch_size, 4, 4) - projection matrices
KRcam = torch.zeros(n_views, batch_size, 4, 4, dtype=torch.float32)
# Set to identity-like projection matrices
for v in range(n_views):
    for b in range(batch_size):
        KRcam[v, b] = torch.eye(4)
        # Add some slight variation for realism
        KRcam[v, b, 0, 0] = 500.0 + v * 10  # focal length variation
        KRcam[v, b, 1, 1] = 500.0 + v * 10
        KRcam[v, b, 0, 2] = w / 2 + b * 5  # principal point
        KRcam[v, b, 1, 2] = h / 2 + b * 5

ARGS = {
    "coords": coords,
    "origin": origin,
    "voxel_size": voxel_size,
    "feats": feats,
    "KRcam": KRcam,
}
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_view_frustum Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zju3dv/NeuralRecon/datasets/transforms.py
code served (permissive licence) · get_code("7458c7475bfd5b04")
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 (3, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Analyze the function signature and body:
# - max_depth: float, used in arithmetic
# - size: tuple of (height, width), unpacked as im_h, im_w
# - cam_intr: tensor with shape (2, 3) accessed at [0, 2], [0, 0], [1, 2], [1, 1]
# - cam_pose: tensor passed to rigid_transform, which expects (N, 4, 4) or similar 4x4 transform matrix

# Build minimal valid arguments:
max_depth = 10.0

size = (480, 640)  # im_h, im_w

# Camera intrinsics matrix: (2, 3) shape with focal lengths and principal point
cam_intr = torch.tensor([
    [500.0, 0.0, 320.0],  # [fx, 0, cx]
    [0.0, 500.0, 240.0]   # [0, fy, cy]
])

# Camera pose: 4x4 rigid transformation matrix (rotation + translation)
# Start with identity and modify slightly
cam_pose = torch.eye(4)
cam_pose[0, 3] = 0.1  # small translation in x
cam_pose[1, 3] = 0.2  # small translation in y
cam_pose[2, 3] = 0.0  # translation in z

ARGS = {
    "max_depth": max_depth,
    "size": size,
    "cam_intr": cam_intr,
    "cam_pose": cam_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 2, dtype family float, shape (3, 5) — shape (3, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
make_nograd_func Ran
ran; our contract draft was wrong, not the code
zju3dv/NeuralRecon/utils.py
code served (permissive licence) · get_code("a76fc7fb6958f97c")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# Create a simple function that uses torch operations
def sample_func(x, y):
    # This function performs some torch operations
    result = x + y
    return result

# Test that the function works with gradients enabled
test_x = torch.randn(2, 4, requires_grad=True)
test_y = torch.randn(2, 4, requires_grad=True)

ARGS = {
    "func": sample_func
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
make_recursive_func Ran
ran; our contract draft was wrong, not the code
zju3dv/NeuralRecon/utils.py
code served (permissive licence) · get_code("b62bec09ceeaca79")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# Define a simple function to be made recursive
def example_func(x):
    """A simple function that doubles its input"""
    if isinstance(x, torch.Tensor):
        return x * 2
    elif isinstance(x, (int, float)):
        return x * 2
    else:
        return x

# The make_recursive_func takes a function as its only parameter
ARGS = {
    "func": example_func
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
pad_scannet Ran
ran; our contract draft was wrong, not the code
zju3dv/NeuralRecon/datasets/transforms.py
code served (permissive licence) · get_code("d9a5fbfa2b1985a1")
How it ran

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

import torch
import numpy as np
from PIL import Image

torch.manual_seed(0)

# Create a PIL Image with ScanNet dimensions (1296x968)
img = Image.new('RGB', (1296, 968), color=(100, 150, 200))

# Create intrinsics matrix (3x3 camera intrinsic matrix)
# The function accesses intrinsics[1, 2], so it needs to be indexable
intrinsics = torch.tensor([
    [500.0, 0.0, 648.0],
    [0.0, 500.0, 484.0],
    [0.0, 0.0, 1.0]
], dtype=torch.float32)

ARGS = {
    'img': img,
    'intrinsics': intrinsics
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
PIL, numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type 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
rigid_transform Ran
ran; our contract draft was wrong, not the code
zju3dv/NeuralRecon/datasets/transforms.py
code served (permissive licence) · get_code("a93502ff16e426c4")
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, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Create a small point cloud: (N, 3) where N=2
xyz = torch.randn(2, 3)

# Create a rigid transform: (4, 4) homogeneous transformation matrix
# This should be a valid 4x4 transformation matrix
transform = torch.eye(4, 4)
transform[0, 3] = 1.0  # translation in x
transform[1, 3] = 2.0  # translation in y
transform[2, 3] = 3.0  # translation in z

ARGS = {
    "xyz": xyz,
    "transform": transform,
}
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, 3) — shape (2, 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
tensor2float Ran
ran; our contract draft was wrong, not the code
zju3dv/NeuralRecon/utils.py
code served (permissive licence) · get_code("25616096f605f638")
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)

# The function tensor2float is decorated with make_recursive_func
# It accepts a single parameter 'vars' which can be:
# - A float (returned as-is)
# - A torch.Tensor (scalar or multi-element, converted to float/list of floats)
# - A list, tuple, or dict (recursively processed)
# - Other types raise NotImplementedError

# Test with a nested structure containing various valid types
scalar_tensor = torch.tensor(3.14)
multi_element_tensor = torch.tensor([1.0, 2.0, 3.0, 4.0])
plain_float = 2.71

ARGS = {
    "vars": {
        "scalar": scalar_tensor,
        "multi": multi_element_tensor,
        "float": plain_float,
        "nested_list": [torch.tensor(1.5), 2.5],
        "nested_tuple": (torch.tensor(5.0), 6.0),
    }
}
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 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

Also present in other papers' code

Of 7 samples lifted from this paper's linked repositories, 7 carry a function identity and 0 do not. Of the 7 distinct functions with an identity, 3 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 4 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
make_nograd_func 16 other papers
3 dated earlier, 13 later
1806.01260 (2018-06)
Digging Into Self-Supervised Monocular Depth Estimation
1912.06378 (2019-12)
Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching
2007.10872 (2020-07)
Dense Hybrid Recurrent Multi-view Stereo Net with Dynamic Consistency Checking
and 13 more not named here
make_recursive_func 14 other papers
2 dated earlier, 12 later
1912.06378 (2019-12)
Cascade Cost Volume for High-Resolution Multi-View Stereo and Stereo Matching
2007.10872 (2020-07)
Dense Hybrid Recurrent Multi-view Stereo Net with Dynamic Consistency Checking
2108.03824 (2021-08)
AA-RMVSNet: Adaptive Aggregation Recurrent Multi-view Stereo Network
and 11 more not named here
tensor2float 1 other paper
1 later
2206.07710 (2022-06)
PlanarRecon: Real-time 3D Plane Detection and Reconstruction from Posed Monocular Videos

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

Repositories linked to this paper

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

Abstract

We present a novel framework named NeuralRecon for real-time 3D scene reconstruction from a monocular video. Unlike previous methods that estimate single-view depth maps separately on each key-frame and fuse them later, we propose to directly reconstruct local surfaces represented as sparse TSDF volumes for each video fragment sequentially by a neural network. A learning-based TSDF fusion module based on gated recurrent units is used to guide the network to fuse features from previous fragments. This design allows the network to capture local smoothness prior and global shape prior of 3D surfaces when sequentially reconstructing the surfaces, resulting in accurate, coherent, and real-time surface reconstruction. The experiments on ScanNet and 7-Scenes datasets show that our system outperforms state-of-the-art methods in terms of both accuracy and speed. To the best of our knowledge, this is the first learning-based system that is able to reconstruct dense coherent 3D geometry in real-time. Code is available at the project page: https://zju3dv.github.io/ neuralrecon/.

For agents

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

get_harvested_code_for_paper("2104.00681")
get_code_for_paper("2104.00681")
have("2104.00681")

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

curl https://syntology.ai/api/ran/2104.00681.json

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

[![Syntology run record](https://syntology.ai/api/ran/2104.00681.svg)](https://syntology.ai/paper/2104.00681)

Connect an agent — have() is free.