SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2304.12294 · 2023

Explicit Correspondence Matching for Generalizable Neural Radiance Fields

arXiv · PDF · Open in the Atlas

Code that ran

8 of 9 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, 1 violated it and 3 have no contract verdict; 4 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/2304.12294.json · badge: /api/ran/2304.12294.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
donydchen/matchnerf canonical (classifier label)
classifier basis: census
8 ran · 1 no recorded run
FunctionStatusWhere it lives
average_poses Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
donydchen/matchnerf/datasets/ibrnet.py
code served (permissive licence) · get_code("e67494c09010d716")
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

# Create sample poses with shape (N_images, 3, 4)
# N_images = 2, 3 rows (x, y, z axes), 4 columns (3 for rotation + 1 for translation)
poses = np.array([
    [[1.0, 0.0, 0.0, 1.0],
     [0.0, 1.0, 0.0, 2.0],
     [0.0, 0.0, 1.0, 3.0]],
    [[0.9, 0.1, 0.0, 1.5],
     [-0.1, 0.9, 0.0, 2.5],
     [0.0, 0.0, 1.0, 3.5]]
], dtype=np.float32)

ARGS = {
    "poses": poses
}
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
center_poses Ran
ran on a synthesized input
donydchen/matchnerf/datasets/ibrnet.py
code served (permissive licence) · get_code("d0d3f4b1601da1e2")
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 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 numpy as np

# Set random seed for reproducibility
np.random.seed(0)

# Create sample poses: (N_images, 3, 4)
# N_images = 2 (small batch)
# Each pose is 3x4: [x_axis, y_axis, z_axis, center]
poses = np.random.randn(2, 3, 4).astype(np.float32)

# Create blender2opencv transformation matrix: (4, 4)
# This is a standard coordinate transformation matrix
blender2opencv = np.array([
    [1, -1, -1, 0],
    [-1, 1, -1, 0],
    [-1, -1, 1, 0],
    [0, 0, 0, 1]
], dtype=np.float32)

ARGS = {
    "poses": poses,
    "blender2opencv": blender2opencv
}
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 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
gen_pairs Ran
ran on a synthesized input
donydchen/matchnerf/datasets/colmap.py
code served (permissive licence) · get_code("7f2d6c9b36d2a638")
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 os
import numpy as np
import tempfile
import shutil

# Create a temporary directory structure with subdirectories containing poses_bounds.npy files
temp_dir = tempfile.mkdtemp()

# Create first scene subdirectory
scene1_dir = os.path.join(temp_dir, 'scene1')
os.makedirs(scene1_dir)

# Create poses_bounds.npy for scene1 with enough images
# poses_bounds shape: (N_images, 17) where first 15 are poses, last 2 are bounds
N_images_1 = 25
poses_bounds_1 = np.random.randn(N_images_1, 17).astype(np.float32)
np.save(os.path.join(scene1_dir, 'poses_bounds.npy'), poses_bounds_1)

# Create second scene subdirectory
scene2_dir = os.path.join(temp_dir, 'scene2')
os.makedirs(scene2_dir)

# Create poses_bounds.npy for scene2 with few images (to test the <=3 branch)
N_images_2 = 3
poses_bounds_2 = np.random.randn(N_images_2, 17).astype(np.float32)
np.save(os.path.join(scene2_dir, 'poses_bounds.npy'), poses_bounds_2)

# Create third scene subdirectory
scene3_dir = os.path.join(temp_dir, 'scene3')
os.makedirs(scene3_dir)

# Create poses_bounds.npy for scene3 with moderate number of images
N_images_3 = 10
poses_bounds_3 = np.random.randn(N_images_3, 17).astype(np.float32)
np.save(os.path.join(scene3_dir, 'poses_bounds.npy'), poses_bounds_3)

ARGS = {
    'root_dir': temp_dir,
    'n_select': 20,
    'n_interval': 6
}
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 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_options Ran
ran; our contract draft was wrong, not the code
donydchen/matchnerf/options.py
code served (permissive licence) · get_code("5dd98ac338acc206")
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 EasyDict. 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 os
import yaml
from easydict import EasyDict as edict

# Create temporary YAML files for testing
temp_dir = tempfile.mkdtemp()

# Create a parent config file
parent_config = {
    'model': {
        'type': 'resnet',
        'layers': 50
    },
    'training': {
        'lr': 0.001,
        'epochs': 100
    }
}

parent_file = os.path.join(temp_dir, 'parent_config.yaml')
with open(parent_file, 'w') as f:
    yaml.dump(parent_config, f)

# Create a child config file that overrides parent
child_config = {
    '_parent_': parent_file,
    'model': {
        'layers': 101
    },
    'training': {
        'lr': 0.0001
    }
}

child_file = os.path.join(temp_dir, 'child_config.yaml')
with open(child_file, 'w') as f:
    yaml.dump(child_config, f)

# Create a simple config file without parent
simple_config = {
    'batch_size': 32,
    'hidden_dim': 256,
    'dropout': 0.5
}

simple_file = os.path.join(temp_dir, 'simple_config.yaml')
with open(simple_file, 'w') as f:
    yaml.dump(simple_config, f)

ARGS = {
    'fname': simple_file
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
easydict, yaml — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EasyDict — a EasyDict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a EasyDict 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
normalize Ran
ran, violated the contract we drafted
donydchen/matchnerf/datasets/ibrnet.py
code served (permissive licence) · get_code("aa33e7122eb26d8b")
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 (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 numpy as np

# Create a simple vector to normalize
v = np.array([3.0, 4.0])

ARGS = {
    "v": v
}
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 (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
parse_arguments Ran
ran on a synthesized input
donydchen/matchnerf/options.py
code served (permissive licence) · get_code("eae59d89065e271f")
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 EasyDict. 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 yaml
from easydict import EasyDict as edict

# The function parse_arguments takes a list of command-line style arguments
# and parses them into a nested dictionary structure using YAML and dot notation.
# It supports:
# - --key.subkey=value  (parsed via YAML)
# - --key.subkey=       (None)
# - --key.subkey        (True)
# - --key.subkey!       (False)
# - --key.subkey=val1,val2,  (list)

ARGS = {
    'args': [
        '--model.name=resnet50',
        '--model.layers=3',
        '--model.pretrained',
        '--training.lr=0.001',
        '--training.epochs=100',
        '--training.disabled!',
        '--data.path=',
        '--augmentation=flip,rotate,crop',
        '--batch_size=32',
        '--verbose',
        '--debug!',
    ]
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
easydict, torch, yaml — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type EasyDict — a EasyDict output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a EasyDict 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
to_hom Ran
ran; our contract draft was wrong, not the code
donydchen/matchnerf/misc/camera.py
code served (permissive licence) · get_code("8ae29d1e7c9ed2f8")
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, 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)

# The function to_hom takes a tensor X and concatenates ones along the last dimension
# X should be a tensor with shape (..., D) where D is some feature dimension
# The function will append a column of ones to make it homogeneous coordinates

X = torch.randn(2, 4)  # batch size 2, feature dimension 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, 5) — shape (2, 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
world2cam Ran
ran; our contract draft was wrong, not the code
donydchen/matchnerf/misc/camera.py
code served (permissive licence) · get_code("7e76538e2f377f4d")
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, 4, 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)

class Pose:
    """
    A class of operations on camera poses (PyTorch tensors with shape [...,3,4])
    each [3,4] camera pose takes the form of [R|t]
    """

    def __call__(self, R=None, t=None):
        assert R is not None or t is not None
        if R is None:
            if not isinstance(t, torch.Tensor):
                t = torch.tensor(t)
            R = torch.eye(3, device=t.device).repeat(*t.shape[:-1], 1, 1)
        elif t is None:
            if not isinstance(R, torch.Tensor):
                R = torch.tensor(R)
            t = torch.zeros(R.shape[:-1], device=R.device)
        else:
            if not isinstance(R, torch.Tensor):
                R = torch.tensor(R)
            if not isinstance(t, torch.Tensor):
                t = torch.tensor(t)
        assert R.shape[:-1] == t.shape and R.shape[-2:] == (3, 3)
        R = R.float()
        t = t.float()
        pose = torch.cat([R, t[..., None]], dim=-1)
        assert pose.shape[-2:] == (3, 4)
        return pose

    def invert(self, pose, use_inverse=False):
        R, t = (pose[..., :3], pose[..., 3:])
        R_inv = R.inverse() if use_inverse else R.transpose(-1, -2)
        t_inv = (-R_inv @ t)[..., 0]
        pose_inv = self(R=R_inv, t=t_inv)
        return pose_inv

    def compose(self, pose_list):
        pose_new = pose_list[0]
        for pose in pose_list[1:]:
            pose_new = self.compose_pair(pose_new, pose)
        return pose_new

    def compose_pair(self, pose_a, pose_b):
        R_a, t_a = (pose_a[..., :3], pose_a[..., 3:])
        R_b, t_b = (pose_b[..., :3], pose_b[..., 3:])
        R_new = R_b @ R_a
        t_new = (R_b @ t_a + t_b)[..., 0]
        pose_new = self(R=R_new, t=t_new)
        return pose_new

def to_hom(X):
    X_hom = torch.cat([X, torch.ones_like(X[..., :1])], dim=-1)
    return X_hom

def world2cam(X, pose):
    X_hom = to_hom(X)
    return X_hom @ pose.transpose(-1, -2)

pose_obj = Pose()
R = torch.eye(3, dtype=torch.float32).unsqueeze(0).repeat(2, 1, 1)
t = torch.zeros(2, 3, dtype=torch.float32)
pose_tensor = pose_obj(R=R, t=t)
X = torch.randn(2, 4, 3, dtype=torch.float32)

ARGS = {
    "X": X,
    "pose": pose_tensor,
}
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, 4, 3) — shape (2, 4, 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
cam2img Not yet run
unverified (harvested, no recorded run)
donydchen/matchnerf/misc/camera.py
code served (permissive licence) · get_code("54e93b2a1fc19f41")
How it ran

no invocation recorded.

Also present in other papers' code

Of 9 samples lifted from this paper's linked repositories, 9 carry a function identity and 0 do not. Of the 9 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, 5 later) and 3 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
normalize 10 other papers
5 dated earlier, 5 later
2008.02268 (2020-08)
NeRF in the Wild: Neural Radiance Fields for Unconstrained Photo Collections
2110.07342 (2021-10)
FILM: Following Instructions in Language with Modular Methods
2205.14870 (2022-05)
Compressible-composable NeRF via Rank-residual Decomposition
and 7 more not named here
to_hom 4 other papers
2 dated earlier, 2 later
2010.10505 (2020-10)
SDF-SRN: Learning Signed Distance 3D Object Reconstruction from Static Images
2104.06405 (2021-04)
BARF : Bundle-Adjusting Neural Radiance Fields
2306.04166 (2023-06)
BAA-NGP: Bundle-Adjusting Accelerated Neural Graphics Primitives
and 1 more not named here
cam2img 3 other papers
2 dated earlier, 1 later
2010.10505 (2020-10)
SDF-SRN: Learning Signed Distance 3D Object Reconstruction from Static Images
2104.06405 (2021-04)
BARF : Bundle-Adjusting Neural Radiance Fields
aaai_27859 (2024 (year only))
Improving Robustness for Joint Optimization of Camera Poses and Decomposed Low-Rank Tensorial Radiance Fields
load_options 3 other papers
2 dated earlier, 1 later
2010.10505 (2020-10)
SDF-SRN: Learning Signed Distance 3D Object Reconstruction from Static Images
2104.06405 (2021-04)
BARF : Bundle-Adjusting Neural Radiance Fields
2312.04565 (2023-12)
MuRF: Multi-Baseline Radiance Fields
world2cam 3 other papers
2 dated earlier, 1 later
2010.10505 (2020-10)
SDF-SRN: Learning Signed Distance 3D Object Reconstruction from Static Images
2104.06405 (2021-04)
BARF : Bundle-Adjusting Neural Radiance Fields
aaai_27859 (2024 (year only))
Improving Robustness for Joint Optimization of Camera Poses and Decomposed Low-Rank Tensorial Radiance Fields
average_poses 1 other paper
1 dated earlier
2008.02268 (2020-08)
NeRF in the Wild: Neural Radiance Fields for Unconstrained Photo Collections

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 new generalizable NeRF method that is able to directly generalize to new unseen scenarios and perform novel view synthesis with as few as two source views. The key to our approach lies in the explicitly modeled correspondence matching information, so as to provide the geometry prior to the prediction of NeRF color and density for volume rendering. The explicit correspondence matching is quantified with the cosine similarity between image features sampled at the 2D projections of a 3D point on different views, which is able to provide reliable cues about the surface geometry. Unlike previous methods where image features are extracted independently for each view, we consider modeling the cross-view interactions via Transformer cross-attention, which greatly improves the feature matching quality. Our method achieves state-of-the-art results on different evaluation settings, with the experiments showing a strong correlation between our learned cosine feature similarity and volume density, demonstrating the effectiveness and superiority of our proposed method. The code and model are on our project page: https://donydchen.github.io/matchnerf

For agents

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

get_harvested_code_for_paper("2304.12294")
get_code_for_paper("2304.12294")
have("2304.12294")

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

curl https://syntology.ai/api/ran/2304.12294.json

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

[![Syntology run record](https://syntology.ai/api/ran/2304.12294.svg)](https://syntology.ai/paper/2304.12294)

Connect an agent — have() is free.