SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2204.13696 · CVPR · 2022

NeurMiPs: Neural Mixture of Planar Experts for View Synthesis

Shenlong Wang, Yu-Chiang Wang, Wei-Chiu Ma, Hao-Yu Hsu, Zhi-Hao Lin

arXiv · PDF · Open in the Atlas

Code that ran

7 of 10 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (4 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 4 have no contract verdict; 3 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/2204.13696.json · badge: /api/ran/2204.13696.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
zhihao-lin/neurmips marked official in Papers with Code
Papers with Code archive
7 ran · 3 no recorded run
FunctionStatusWhere it lives
Experts Ran
ran on a synthesized input · constructed an object; no result was computed
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("664deb75e8d8b7af")
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, Experts: 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 math

torch.manual_seed(0)

# Construct valid arguments for Experts.__init__
ARGS = {
    "n_experts": 3,
    "in_features": 4,
    "out_features": 5
}
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 Experts — 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 Experts 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
HarmonicEmbedding Ran
ran on a synthesized input · constructed an object; no result was computed
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("2151fb3a734dc215")
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, HarmonicEmbedding: 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)

# The HarmonicEmbedding.__init__ takes three parameters:
# - n_harmonic_functions: int (default 6)
# - omega0: float (default 1.0)
# - logspace: bool (default True)

ARGS = {
    'n_harmonic_functions': 4,
    'omega0': 2.0,
    'logspace': True,
}
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 HarmonicEmbedding — 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 HarmonicEmbedding 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
MLPWithInputSkips Ran
ran on a synthesized input · constructed an object; no result was computed
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("b141a7d809fb861a")
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, MLPWithInputSkips: 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 math

torch.manual_seed(0)

ARGS = {
    "n_experts": 3,
    "n_layers": 4,
    "input_dim": 5,
    "output_dim": 6,
    "skip_dim": 7,
    "hidden_dim": 8,
    "input_skips": (1, 2),
}
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 MLPWithInputSkips — 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 MLPWithInputSkips 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
NerfExperts Ran
ran on a synthesized input · constructed an object; no result was computed
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("8bbe49633a04bd6e")
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, NerfExperts: 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 math

torch.manual_seed(0)

class Experts(nn.Module):
    """
    Single MLP layer with multiple experts
    """

    def __init__(self, n_experts: int, in_features: int, out_features: int):
        super().__init__()
        self.n_experts = n_experts
        self.weight = nn.Parameter(torch.empty(n_experts, in_features, out_features))
        self.bias = nn.Parameter(torch.empty(n_experts, out_features))
        self.reset_parameters()

    def reset_parameters(self):
        for i in range(self.n_experts):
            nn.init.xavier_uniform_(self.weight[i])
        fan_in = self.weight.shape[1]
        bound = 1 / math.sqrt(fan_in)
        nn.init.uniform_(self.bias, -bound, bound)

    def forward(self, inputs, index):
        """
        Args
            inputs: (b, in_feat)
            index: (b, ), max < n_experts
        Return 
            outputs: (b, out_feat)
        """
        weight_sample = self.weight[index]
        bias_sample = self.bias[index]
        prod = torch.einsum('...i,...io->...o', inputs, weight_sample)
        outputs = prod + bias_sample
        return outputs


class HarmonicEmbedding(torch.nn.Module):

    def __init__(self, n_harmonic_functions: int=6, omega0: float=1.0, logspace: bool=True):
        super().__init__()
        if logspace:
            frequencies = 2.0 ** torch.arange(n_harmonic_functions, dtype=torch.float32)
        else:
            frequencies = torch.linspace(1.0, 2.0 ** (n_harmonic_functions - 1), n_harmonic_functions, dtype=torch.float32)
        self.register_buffer('_frequencies', omega0 * frequencies)

    def forward(self, x: torch.Tensor):
        """
        Args:
            x: tensor of shape [..., dim]
        Returns:
            embedding: a harmonic embedding of `x`
                of shape [..., n_harmonic_functions * dim * 2]
        """
        embed = (x[..., None] * self._frequencies).view(*x.shape[:-1], -1)
        return torch.cat((embed.sin(), embed.cos()), dim=-1)


class MLPWithInputSkips(nn.Module):

    def __init__(self, n_experts: int, n_layers: int, input_dim: int, output_dim: int, skip_dim: int, hidden_dim: int, input_skips=None):
        if input_skips is None:
            input_skips = []
        super().__init__()
        layers = []
        for layeri in range(n_layers):
            if layeri == 0:
                dimin = input_dim
                dimout = hidden_dim
            elif layeri in input_skips:
                dimin = hidden_dim + skip_dim
                dimout = hidden_dim
            else:
                dimin = hidden_dim
                dimout = output_dim
            experts = Experts(n_experts, dimin, dimout)
            layers.append(experts)
        self.layers = torch.nn.ModuleList(layers)
        self._input_skips = set(input_skips)

    def forward(self, x, z, indices):
        """
        Args:
            x: input tensor of shape (batch, input_dim)
            z: skip tensor of shape (batch, skip_dim)
            indices: expert indices for each layer, list of (batch,) tensors
        Returns:
            output tensor of shape (batch, output_dim)
        """
        out = x
        for i, layer in enumerate(self.layers):
            if i in self._input_skips:
                out = torch.cat([out, z], dim=-1)
            out = layer(out, indices[i])
        return out


# Construct test arguments
n_experts = 3
n_layers = 2
input_dim = 4
output_dim = 5
skip_dim = 6
hidden_dim = 8
batch_size = 2

# Create input tensors
x = torch.randn(batch_size, input_dim)
z = torch.randn(batch_size, skip_dim)

# Create expert indices for each layer
indices = [torch.randint(0, n_experts, (batch_size,)) for _ in range(n_layers)]

# Create the model
model = MLPWithInputSkips(
    n_experts=n_experts,
    n_layers=n_layers,
    input_dim=input_dim,
    output_dim=output_dim,
    skip_dim=skip_dim,
    hidden_dim=hidden_dim,
    input_skips=[1]
)

ARGS = {
    'x': x,
    'z': z,
    'indices': indices,
}
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 NerfExperts — 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 NerfExperts 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
farthest_point_sample Ran
ran; our contract draft was wrong, not the code
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("682c586fc4728495")
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

torch.manual_seed(0)

# Create a small point cloud tensor
# Shape: (point_n=5, dim=3)
points = torch.randn(5, 3)

# Number of points to sample
sample_n = 3

ARGS = {
    "points": points,
    "sample_n": sample_n,
}
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_points_lrf Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("19ee74bf02bae244")
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, 3, 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
import math

torch.manual_seed(0)

# Create small test tensors
# points: (point_n, 3) - 5 points in 3D space
points = torch.randn(5, 3)

# indices: (sample_n,) - indices of points to sample from
# Let's select 2 points out of 5
indices = torch.tensor([0, 2])

# neighbor_num: number of neighbors to consider
neighbor_num = 3

# chunk_size: processing chunk size
chunk_size = 2

ARGS = {
    "points": points,
    "neighbor_num": neighbor_num,
    "indices": indices,
    "chunk_size": chunk_size,
}
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, 3, 3) — shape (2, 3, 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
orthonormal_basis_from_xy Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("2fa5b9bc82100a84")
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, 3, 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
import torch.nn.functional as F

torch.manual_seed(0)

# The function expects xy to be a tensor of shape (n, 3, 2)
# where n is batch size, 3 is the number of 3D vectors, and 2 represents x,y components
# We need to create a valid input tensor

# Create a batch of 2 samples, each with 3 vectors in 2D (x, y components)
# Shape: (batch_size=2, num_vectors=3, xy_components=2)
xy = torch.randn(2, 3, 2)

ARGS = {
    "xy": xy
}
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, 3, 3) — shape (2, 3, 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
ModelExperts Not yet run
unverified (harvested, no recorded run)
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("71e417ec8a9d5f9a")
How it ran

no invocation recorded.

PlaneGeometry Not yet run
unverified (harvested, no recorded run)
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("2c5818521dbcd604")
How it ran

no invocation recorded.

_xavier_init Not yet run
unverified (harvested, no recorded run)
zhihao-lin/neurmips/mnh/model_experts.py
code served (permissive licence) · get_code("00e34de851ece206")
How it ran

no invocation recorded.

Also present in other papers' code

Of 10 samples lifted from this paper's linked repositories, 10 carry a function identity and 0 do not. Of the 10 distinct functions with an identity, 0 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 0 later) and 10 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.

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

Figure 1. Comparison between different 3D representations for neural rendering. Neural implicit surface models accurate surface geometry but rendering requires expensive sequential sampling. Multi-planar imagery is efficient, but does not reflect true geometry and is not suitable for extrapolation. While NeRF is flexible, it is not sample-efficient and its learned density field might not reflect true scene geometry. Our proposed Mixture of Planar Experts is efficient and is able to model complicated surface geometry and appearance.

For agents

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

get_harvested_code_for_paper("2204.13696")
get_code_for_paper("2204.13696")
have("2204.13696")

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

curl https://syntology.ai/api/ran/2204.13696.json

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

[![Syntology run record](https://syntology.ai/api/ran/2204.13696.svg)](https://syntology.ai/paper/2204.13696)

Connect an agent — have() is free.