SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2305.15399 · ICLR · 2024

Sin3DM: Learning a Diffusion Model from a Single 3D Textured Shape

Rundi Wu, Carl Vondrick, Changxi Zheng, Ruoshi Liu

arXiv · PDF · Open in the Atlas

Code that ran

8 of 10 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (2 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 6 have no contract verdict; 2 failures of our instrument; 2 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/2305.15399.json · badge: /api/ran/2305.15399.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
sin3dm/sin3dm marked official in Papers with Code
Papers with Code archive
8 ran · 2 no recorded run
FunctionStatusWhere it lives
SiLU Ran
ran on a synthesized input
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("5465fa33e2f8631e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) 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 as th
import torch.nn as nn

class SiLU(nn.Module):

    def forward(self, x):
        return x * th.sigmoid(x)


INIT_ARGS = {}

ARGS = {
    'x': th.randn(2, 4)
}

if __name__ == '__main__':
    model = SiLU(**INIT_ARGS)
    output = model(**ARGS)
    print(output)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
TriplaneConv Ran
ran on a synthesized input · constructed an object; no result was computed
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("ffc71c22830d5c71")
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, TriplaneConv: 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 as th
import torch.nn as nn

th.manual_seed(0)

# Parameters for TriplaneConv.__init__
channels = 4
out_channels = 8
kernel_size = 3
padding = 1
is_rollout = True

# Create the module
module = nn.Module()

# Create sample input tensors for the forward pass
# The forward expects a tuple of 3 feature maps (xy, xz, yz planes)
# Each should have shape (batch, channels, height, width) or (batch, channels, height, depth)
batch_size = 2
H = 5  # height for xy and xz planes
W = 6  # width for xy and yz planes
D = 7  # depth for xz and yz planes

tpl_xy = th.randn(batch_size, channels, H, W)
tpl_xz = th.randn(batch_size, channels, H, D)
tpl_yz = th.randn(batch_size, channels, W, D)

featmaps = (tpl_xy, tpl_xz, tpl_yz)

ARGS = {
    "channels": channels,
    "out_channels": out_channels,
    "kernel_size": kernel_size,
    "padding": padding,
    "is_rollout": is_rollout,
}
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 TriplaneConv — 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 TriplaneConv 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
TriplaneDownsample2x Ran
ran on a synthesized input
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("b722a7e8490748a1")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) 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 torch.nn as nn
import torch.nn.functional as F

class TriplaneDownsample2x(nn.Module):

    def __init__(self) -> None:
        super().__init__()

    def forward(self, featmaps):
        tpl_xy, tpl_xz, tpl_yz = featmaps
        H, W = tpl_xy.shape[-2:]
        D = tpl_xz.shape[-1]
        tpl_xy = F.avg_pool2d(tpl_xy, kernel_size=2, stride=2)
        tpl_xz = F.avg_pool2d(tpl_xz, kernel_size=2, stride=2)
        tpl_yz = F.avg_pool2d(tpl_yz, kernel_size=2, stride=2)
        assert tpl_xy.shape[-2] == H // 2 and tpl_xy.shape[-1] == W // 2
        assert tpl_xz.shape[-2] == H // 2 and tpl_xz.shape[-1] == D // 2
        assert tpl_yz.shape[-2] == W // 2 and tpl_yz.shape[-1] == D // 2
        return (tpl_xy, tpl_xz, tpl_yz)


INIT_ARGS = {}

ARGS = {
    'featmaps': (
        torch.randn(2, 4, 8, 6),
        torch.randn(2, 4, 8, 5),
        torch.randn(2, 4, 6, 5)
    )
}

if __name__ == '__main__':
    model = TriplaneDownsample2x(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shapes:")
    print(f"  tpl_xy: {output[0].shape}")
    print(f"  tpl_xz: {output[1].shape}")
    print(f"  tpl_yz: {output[2].shape}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
TriplaneNorm Ran
ran on a synthesized input · constructed an object; no result was computed
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("af4e0bdd202a4062")
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, TriplaneNorm: 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)

class GroupNorm32(nn.GroupNorm):
    def forward(self, x):
        return super().forward(x.float()).type(x.dtype)

def normalization(channels):
    return GroupNorm32(32, channels)

class TriplaneNorm(nn.Module):
    def __init__(self, channels) -> None:
        super().__init__()
        self.norm_xy = normalization(channels)
        self.norm_xz = normalization(channels)
        self.norm_yz = normalization(channels)

    def forward(self, featmaps):
        tpl_xy, tpl_xz, tpl_yz = featmaps
        H, W = tpl_xy.shape[-2:]
        D = tpl_xz.shape[-1]
        tpl_xy_h = self.norm_xy(tpl_xy)
        tpl_xz_h = self.norm_xz(tpl_xz)
        tpl_yz_h = self.norm_yz(tpl_yz)
        assert tpl_xy_h.shape[-2] == H and tpl_xy_h.shape[-1] == W
        assert tpl_xz_h.shape[-2] == H and tpl_xz_h.shape[-1] == D
        assert tpl_yz_h.shape[-2] == W and tpl_yz_h.shape[-1] == D
        return (tpl_xy_h, tpl_xz_h, tpl_yz_h)

channels = 64
H = 4
W = 3
D = 5

tpl_xy = torch.randn(2, channels, H, W)
tpl_xz = torch.randn(2, channels, H, D)
tpl_yz = torch.randn(2, channels, W, D)

featmaps = (tpl_xy, tpl_xz, tpl_yz)

ARGS = {
    "channels": channels
}
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 TriplaneNorm — 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 TriplaneNorm 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
TriplaneSiLU Ran
ran on a synthesized input
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("462d70c6887d1909")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) 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 as th
import torch.nn as nn

class SiLU(nn.Module):

    def forward(self, x):
        return x * th.sigmoid(x)

class TriplaneSiLU(nn.Module):

    def __init__(self) -> None:
        super().__init__()
        self.silu = SiLU()

    def forward(self, featmaps):
        tpl_xy, tpl_xz, tpl_yz = featmaps
        return (self.silu(tpl_xy), self.silu(tpl_xz), self.silu(tpl_yz))


INIT_ARGS = {}

ARGS = {
    'featmaps': (
        th.randn(2, 4, 3),
        th.randn(2, 4, 5),
        th.randn(2, 4, 7)
    )
}

if __name__ == '__main__':
    model = TriplaneSiLU(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shapes:", tuple(o.shape for o in output))
    print("Success!")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
TriplaneUpsample2x Ran
ran on a synthesized input
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("4d7539fb91db4193")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) 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 torch.nn as nn
import torch.nn.functional as F

class TriplaneUpsample2x(nn.Module):

    def __init__(self) -> None:
        super().__init__()

    def forward(self, featmaps):
        tpl_xy, tpl_xz, tpl_yz = featmaps
        H, W = tpl_xy.shape[-2:]
        D = tpl_xz.shape[-1]
        tpl_xy = F.interpolate(tpl_xy, scale_factor=2, mode='bilinear', align_corners=False)
        tpl_xz = F.interpolate(tpl_xz, scale_factor=2, mode='bilinear', align_corners=False)
        tpl_yz = F.interpolate(tpl_yz, scale_factor=2, mode='bilinear', align_corners=False)
        assert tpl_xy.shape[-2] == H * 2 and tpl_xy.shape[-1] == W * 2
        assert tpl_xz.shape[-2] == H * 2 and tpl_xz.shape[-1] == D * 2
        assert tpl_yz.shape[-2] == W * 2 and tpl_yz.shape[-1] == D * 2
        return (tpl_xy, tpl_xz, tpl_yz)


INIT_ARGS = {}

ARGS = {
    'featmaps': (
        torch.randn(2, 4, 8, 6),
        torch.randn(2, 4, 8, 5),
        torch.randn(2, 4, 6, 5),
    )
}

if __name__ == '__main__':
    model = TriplaneUpsample2x(**INIT_ARGS)
    output = model(**ARGS)
    print("Output shapes:")
    print(f"  tpl_xy: {output[0].shape}")
    print(f"  tpl_xz: {output[1].shape}")
    print(f"  tpl_yz: {output[2].shape}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
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: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
compose_featmaps Ran
ran; our contract draft was wrong, not the code
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("a025e8940ec65262")
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 feature maps with distinct sizes for different dimensions
# feat_xy: shape [batch, channels, H, W]
# feat_xz: shape [batch, channels, H, D]
# feat_yz: shape [batch, channels, W, D]

batch_size = 2
channels = 4
H = 3
W = 5
D = 7

feat_xy = torch.randn(batch_size, channels, H, W, dtype=torch.float32)
feat_xz = torch.randn(batch_size, channels, H, D, dtype=torch.float32)
feat_yz = torch.randn(batch_size, channels, W, D, dtype=torch.float32)

ARGS = {
    "feat_xy": feat_xy,
    "feat_xz": feat_xz,
    "feat_yz": feat_yz,
}
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
decompose_featmaps Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("662f534ffa6ba3e6")
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)

# The function decomposes a composed feature map into three 2D projections
# by slicing along spatial dimensions and transposing one of them.
# 
# Parameters:
# - composed_map: A tensor that is sliced as [..., :H, :W], [..., :H, W:], [..., H:, :W]
#   The function expects the last two dimensions to be spatial (height and width of composed map)
#   Shape should be compatible with indexing and transposing
# - sizes: A tuple/list of (H, W, D) where H and W are used for slicing

# Create sizes
H, W, D = 4, 5, 3
sizes = (H, W, D)

# Create composed_map tensor
# It needs to have at least (H + D) rows and (W + W) = (2*W) columns in the last two dimensions
# The composed map concatenates three feature maps: xy (H x W), xz (H x W), yz (D x W)
# So the composed map has shape [..., max(H, D), max(W, W) + W] = [..., max(4,3), 5+5] = [..., 4, 10]
# or more precisely: [..., H+D, 2*W] to accommodate all slices
# Let's use batch size 2 and feature dim 4
batch_size = 2
feat_dim = 4
composed_map = torch.randn(batch_size, feat_dim, H + D, 2 * W)

ARGS = {
    "composed_map": composed_map,
    "sizes": sizes,
}
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
TriplaneResBlock Not yet run
unverified (harvested, no recorded run)
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("c2223fab7e4c14ea")
How it ran

no invocation recorded.

TriplaneUNetModelSmall Not yet run
unverified (harvested, no recorded run)
sin3dm/sin3dm/src/diffusion/unet_triplane.py
code served (permissive licence) · get_code("a301d2e57e060921")
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, 3 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 1 the same month, 3 later) and 7 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
SiLU
brick of this paper's assembly: helper
7 other papers
1 dated earlier, 1 the same month, 5 later · 6 only as an assembly brick
2006.11239 (2020-06)
Denoising Diffusion Probabilistic Models
2305.19094 (2023-05)
Diffusion Model for Dense Matching
2310.15587 (2023-10)
SCANDL: A Diffusion Model for Generating Synthetic Scanpaths on Texts
and 4 more not named here
TriplaneNorm
brick of this paper's assembly: model
1 other paper
1 later · 1 only as an assembly brick
2403.07773 (2024-03)
SemCity: Semantic Scene Generation with Triplane Diffusion
TriplaneSiLU
brick of this paper's assembly: model
1 other paper
1 later · 1 only as an assembly brick
2403.07773 (2024-03)
SemCity: Semantic Scene Generation with Triplane Diffusion

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

Synthesizing novel 3D models that resemble the input example has long been pursued by graphics artists and machine learning researchers. In this paper, we present Sin3DM, a diffusion model that learns the internal patch distribution from a single 3D textured shape and generates high-quality variations with fine geometry and texture details. Training a diffusion model directly in 3D would induce large memory and computational cost. Therefore, we first compress the input into a lower-dimensional latent space and then train a diffusion model on it. Specifically, we encode the input 3D textured shape into triplane feature maps that represent the signed distance and texture fields of the input. The denoising network of our diffusion model has a limited receptive field to avoid overfitting, and uses triplane-aware 2D convolution blocks to improve the result quality. Aside from randomly generating new samples, our model also facilitates applications such as retargeting, outpainting and local editing. Through extensive qualitative and quantitative evaluation, we show that our method outperforms prior methods in generation quality of 3D shapes.

For agents

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

get_harvested_code_for_paper("2305.15399")
get_code_for_paper("2305.15399")
have("2305.15399")

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

curl https://syntology.ai/api/ran/2305.15399.json

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

[![Syntology run record](https://syntology.ai/api/ran/2305.15399.svg)](https://syntology.ai/paper/2305.15399)

Connect an agent — have() is free.