SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.14435 · CVPR · 2023

NeRF-DS: Neural Radiance Fields for Dynamic Specular Objects

Gim Lee, Chen Li, Zhiwen Yan

arXiv · PDF · Open in the Atlas

Code that ran

10 of 11 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 7 have no contract verdict; 2 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/2303.14435.json · badge: /api/ran/2303.14435.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
JokerYan/NeRF-DS canonical (classifier label)
classifier basis not recorded on the link
10 ran · 1 no recorded run
FunctionStatusWhere it lives
cal_weights Ran
ran on a synthesized input
JokerYan/NeRF-DS/hypernerf/model_utils.py
code served (permissive licence) · get_code("038164f235e3395a")
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 ArrayImpl, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import jax.numpy as jnp

# Create small test tensors
# sigma: shape (batch, num_samples) - standard deviation values
sigma = jnp.array([[0.5, 1.0, 0.8, 1.2],
                   [0.9, 0.7, 1.1, 0.6]])

# z_vals: shape (batch, num_samples) - depth values along rays
z_vals = jnp.array([[1.0, 2.0, 3.0, 4.0],
                    [1.5, 2.5, 3.5, 4.5]])

# dirs: shape (batch, 3) - ray directions (normalized)
dirs = jnp.array([[0.577, 0.577, 0.577],
                  [0.707, 0.0, 0.707]])

# Optional parameters with defaults
sample_at_infinity = True
eps = 1e-10
scale = 1

ARGS = {
    "sigma": sigma,
    "z_vals": z_vals,
    "dirs": dirs,
    "sample_at_infinity": sample_at_infinity,
    "eps": eps,
    "scale": scale,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
dual_part Ran
ran on a synthesized input
JokerYan/NeRF-DS/hypernerf/dual_quaternion.py
code served (permissive licence) · get_code("edf1956cf4a5038b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# A dual quaternion is typically represented as an 8-element vector
# [w, x, y, z, dw, dx, dy, dz] where the first 4 are the real part
# and the last 4 are the dual part.
# The function extracts elements from index 4 onwards (the dual part).

# Create a batch of dual quaternions with shape (batch_size, 8)
# where batch_size=2 for a small example
dq = torch.randn(2, 8)

ARGS = {
    "dq": dq
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
encode_metadata Ran
ran; our contract draft was wrong, not the code
JokerYan/NeRF-DS/hypernerf/evaluation.py
code served (permissive licence) · get_code("2ea4bc48ba7c9db0")
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
import torch.nn as nn

torch.manual_seed(0)

class MockNerfModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.use_nerf_embed = True
        self.use_warp = True
        self.has_hyper_embed = True
        self.linear = nn.Linear(4, 4)
    
    def encode_nerf_embed(self, metadata):
        return torch.randn(2, 4)
    
    def encode_warp_embed(self, metadata):
        return torch.randn(2, 4)
    
    def encode_hyper_embed(self, metadata):
        return torch.randn(2, 4)
    
    def apply(self, variables, metadata, method=None):
        if method is not None:
            return method(metadata)
        return metadata

model = MockNerfModel()

params = {
    'nerf_embed': torch.randn(3, 4),
    'warp_embed': torch.randn(2, 5),
    'hyper_embed': torch.randn(2, 6)
}

metadata = {
    'frame_idx': torch.tensor([0, 1]),
    'camera_idx': torch.tensor([2, 3]),
    'extra': torch.randn(2, 4)
}

ARGS = {
    'model': model,
    'params': params,
    'metadata': metadata
}
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
filter_sigma Ran
ran, violated the contract we drafted
JokerYan/NeRF-DS/hypernerf/tmp_nerf.py
code served (permissive licence) · get_code("5759498c262e96a8")
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 ArrayImpl, ndim 2, dtype family float, shape (2, 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 jax.numpy as jnp
import numpy as np

# Construct arguments for filter_sigma
points = jnp.array([
    [[0.5, 0.5, 0.5], [1.5, 1.5, 1.5]],
    [[0.2, 0.2, 0.2], [2.5, 2.5, 2.5]]
], dtype=jnp.float32)

sigma = jnp.array([
    [0.8, 0.1],
    [0.05, 0.9]
], dtype=jnp.float32)

render_opts = {
    'dust_threshold': 0.15,
    'bounding_box': (0.0, 2.0, 0.0, 2.0, 0.0, 2.0)
}

ARGS = {
    'points': points,
    'sigma': sigma,
    'render_opts': render_opts
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, numpy — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 2, dtype family float, shape (2, 2) — shape (2, 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
get_norm_layer Ran
ran on a synthesized input
JokerYan/NeRF-DS/hypernerf/modules.py
code served (permissive licence) · get_code("454ccbcc18d14a42")
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 partial. 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 functools
from flax import linen as nn

# Test different norm_type values
ARGS = {
    'norm_type': 'layer'
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
flax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type partial — a partial output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a partial 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
real_part Ran
ran on a synthesized input
JokerYan/NeRF-DS/hypernerf/dual_quaternion.py
code served (permissive licence) · get_code("8226212df4de9c5e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# The function real_part(dq) expects dq to be indexable with [..., :4]
# This means it should be a tensor with at least one dimension, where the last dimension has size >= 4
# A dual quaternion typically has 8 components (4 for real part, 4 for dual part)
# We'll create a small batch of dual quaternions

dq = torch.randn(2, 8)

ARGS = {
    "dq": dq
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
sample_along_rays Ran
ran; our contract draft was wrong, not the code
JokerYan/NeRF-DS/hypernerf/model_utils.py
code served (permissive licence) · get_code("6ac3f9fa2286793a")
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 jax
import jax.numpy as jnp
from jax import random

# Set seed for reproducibility
key = random.PRNGKey(0)

# Define tiny tensors for testing
batch_size = 2
num_coarse_samples = 4
origin_dim = 3
direction_dim = 3

# Ray origins: [batch_size, 3]
origins = jnp.array([
    [0.0, 0.0, 0.0],
    [1.0, 1.0, 1.0]
], dtype=jnp.float32)

# Ray directions: [batch_size, 3]
directions = jnp.array([
    [0.0, 0.0, 1.0],
    [0.0, 0.0, 1.0]
], dtype=jnp.float32)

# Near and far clip planes
near = 0.1
far = 10.0

ARGS = {
    "key": key,
    "origins": origins,
    "directions": directions,
    "num_coarse_samples": num_coarse_samples,
    "near": near,
    "far": far,
    "use_stratified_sampling": True,
    "use_linear_disparity": False
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax — 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
sort_camera_paths Ran
ran on a synthesized input
JokerYan/NeRF-DS/render.py
code served (permissive licence) · get_code("1c464999c646d8f5")
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 list, ndim 1, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

from pathlib import Path

# Create mock Path objects with stem attributes
camera_paths = [
    Path("camera_0_front.txt"),
    Path("camera_2_back.txt"),
    Path("camera_1_side.txt"),
]

ARGS = {
    "camera_paths": camera_paths,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
none outside the standard library — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type list, ndim 1, dtype family float — a list 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
split_parts Ran
ran on a synthesized input
JokerYan/NeRF-DS/hypernerf/dual_quaternion.py
code served (permissive licence) · get_code("8694f14e6351b246")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned tuple, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# A dual quaternion is represented as an 8-dimensional vector
# where the first 4 components are the real part (quaternion)
# and the last 4 components are the dual part (quaternion)
dq = torch.randn(2, 8)

ARGS = {
    "dq": dq
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 3, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
volumetric_rendering Ran
ran on a synthesized input
JokerYan/NeRF-DS/hypernerf/model_utils.py
code served (permissive licence) · get_code("a046b1211e508f53")
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 jax.numpy as jnp
import torch

torch.manual_seed(0)

# Create small test tensors
batch_size = 2
num_samples = 4
rgb_dim = 3
dir_dim = 3

# rgb: (B, S, 3) - batch_size=2, num_samples=4, channels=3
rgb = jnp.array(torch.randn(batch_size, num_samples, rgb_dim).numpy())

# sigma: (B, S) - batch_size=2, num_samples=4
sigma = jnp.array(torch.randn(batch_size, num_samples).numpy().clip(0, 1))

# z_vals: (B, S) - batch_size=2, num_samples=4
z_vals = jnp.array(torch.linspace(0.1, 10, num_samples).unsqueeze(0).expand(batch_size, -1).numpy())

# dirs: (B, 3) - batch_size=2, direction vectors
dirs = jnp.array(torch.randn(batch_size, dir_dim).numpy())
# Normalize directions
dirs = dirs / jnp.linalg.norm(dirs, axis=-1, keepdims=True)

# use_white_background: boolean flag
use_white_background = True

# sample_at_infinity: boolean flag
sample_at_infinity = True

# eps: small number for numerical stability
eps = 1e-10

# use_sharp_weights: boolean flag
use_sharp_weights = False

# sharp_weights_std: standard deviation for sharpening
sharp_weights_std = 1.0

ARGS = {
    'rgb': rgb,
    'sigma': sigma,
    'z_vals': z_vals,
    'dirs': dirs,
    'use_white_background': use_white_background,
    'sample_at_infinity': sample_at_infinity,
    'eps': eps,
    'use_sharp_weights': use_sharp_weights,
    'sharp_weights_std': sharp_weights_std,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
jax, 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
get_bone_probs_batch Not yet run
unverified (harvested, no recorded run)
JokerYan/NeRF-DS/hypernerf/bone_utils.py
code served (permissive licence) · get_code("8731bc04d825f433")
How it ran

no invocation recorded.

Also present in other papers' code

Of 11 samples lifted from this paper's linked repositories, 11 carry a function identity and 0 do not. Of the 11 distinct functions with an identity, 7 are also present in other papers whose code we hold (7 with at least one paper dated earlier, 0 the same month, 0 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
filter_sigma 2 other papers
2 dated earlier
2106.13228 (2021-06)
HyperNeRF: A Higher-Dimensional Representation for Topologically Varying Neural Radiance Fields
2112.01983 (2021-12)
CoNeRF: Controllable Neural Radiance Fields
dual_part 1 other paper
1 dated earlier
2112.01983 (2021-12)
CoNeRF: Controllable Neural Radiance Fields
encode_metadata 1 other paper
1 dated earlier
2112.01983 (2021-12)
CoNeRF: Controllable Neural Radiance Fields
get_norm_layer 1 other paper
1 dated earlier
2112.01983 (2021-12)
CoNeRF: Controllable Neural Radiance Fields
real_part 1 other paper
1 dated earlier
2112.01983 (2021-12)
CoNeRF: Controllable Neural Radiance Fields
sample_along_rays 1 other paper
1 dated earlier
2112.01983 (2021-12)
CoNeRF: Controllable Neural Radiance Fields
split_parts 1 other paper
1 dated earlier
2112.01983 (2021-12)
CoNeRF: Controllable Neural Radiance Fields

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

Dynamic Neural Radiance Field (NeRF) is a powerful algorithm capable of rendering photo-realistic novel view images from a monocular RGB video of a dynamic scene. Although it warps moving points across frames from the observation spaces to a common canonical space for rendering, dynamic NeRF does not model the change of the reflected color during the warping. As a result, this approach often fails drastically on challenging specular objects in motion. We address this limitation by reformulating the neural radiance field function to be conditioned on surface position and orientation in the observation space. This allows the specular surface at different poses to keep the different reflected colors when mapped to the common canonical space. Additionally, we add the mask of moving objects to guide the deformation field. As the specular surface changes color during motion, the mask mitigates the problem of failure to find temporal correspondences with only RGB supervision. We evaluate our model based on the novel view synthesis quality with a self-collected dataset of different moving specular objects in realistic environments. The experimental results demonstrate that our method significantly improves the reconstruction quality of moving specular objects from monocular RGB videos compared to the existing NeRF models. Our code and data are available at the project website 1 .

For agents

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

get_harvested_code_for_paper("2303.14435")
get_code_for_paper("2303.14435")
have("2303.14435")

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

curl https://syntology.ai/api/ran/2303.14435.json

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

[![Syntology run record](https://syntology.ai/api/ran/2303.14435.svg)](https://syntology.ai/paper/2303.14435)

Connect an agent — have() is free.