SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 1912.05237 · 2019

Towards Unsupervised Learning of Generative Models for 3D Controllable Image Synthesis

arXiv · PDF · Open in the Atlas

Code that ran

4 of 4 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (0 of them constructed an object rather than computing a result; 0 honoured the contract we drafted, 0 violated it and 3 have no contract verdict; 1 failure of our instrument; 0 no recorded run), measured 2026-11-22; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/1912.05237.json · badge: /api/ran/1912.05237.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
autonomousvision/controllable_image_synthesis canonical (classifier label)
classifier basis: census
4 ran · 0 no recorded run
FunctionStatusWhere it lives
actvn Ran
ran; our contract draft was wrong, not the code
autonomousvision/controllable_image_synthesis/controllable_gan/models/generator_2d.py
code served (permissive licence) · get_code("5bd96fb5938773d4")
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 torch.nn.functional as F

torch.manual_seed(0)

# The function actvn takes a single parameter x and applies leaky_relu to it
# x should be a tensor
x = torch.randn(2, 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, 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
dis2lineseg Ran
ran on a synthesized input
autonomousvision/controllable_image_synthesis/controllable_gan/render/square.py
code served (permissive licence) · get_code("5a9ffd330cb523fd")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API) and returned Tensor, ndim 2, dtype family float, shape (2, 3) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch

torch.manual_seed(0)

# Construct arguments for dis2lineseg
# The function expects:
# - p: point(s) to measure distance from
# - cross: cross product or perpendicular vector(s)
# - c0: first endpoint of line segment
# - c1: second endpoint of line segment

# All tensors need to support operations along dim=2, so they should be 3D
# Shape: (batch, num_points, feature_dim)

batch_size = 2
num_points = 3
feature_dim = 4

# Create point tensor (batch_size, num_points, feature_dim)
p = torch.randn(batch_size, num_points, feature_dim)

# Create cross product tensor (batch_size, num_points, feature_dim)
cross = torch.randn(batch_size, num_points, feature_dim)

# Create line segment endpoints - same shape as p
c0 = torch.randn(batch_size, num_points, feature_dim)
c1 = torch.randn(batch_size, num_points, feature_dim)

ARGS = {
    "p": p,
    "cross": cross,
    "c0": c0,
    "c1": c1
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (2, 3) — shape (2, 3) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
project_cuboids Ran
ran on a synthesized input
autonomousvision/controllable_image_synthesis/controllable_gan/render/square.py
code served (permissive licence) · get_code("c9469018b548dd02")
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)

# Analyze the function signature and body:
# project_cuboids(normals, corners, rays, quad_feat, quad2cam, scale, sigma=1e-05, gamma=0.0001, near_plane=0.0, far_plane=None)
#
# From the docstring and usage:
# - normals: BxNx3 (normal vectors)
# - corners: BxNx4x3 (vertices of quadmeshes)
# - rays: HxWx3 (camera rays)
# - quad_feat: BxNxCxhxw (quadmesh features)
# - quad2cam: BxNx4x4 (transformation matrices)
# - scale: BxNx2 (scale of quadmeshes)
# - sigma, gamma, near_plane, far_plane: scalars/optional
#
# The function calls project_faces in a loop (implied by the structure)
# project_faces expects:
#   n: Nx3, p0: Nx3, c: Nx4x3, l: HxWx3, featmap: NxCxHxW, 
#   quad2cam: Nx4x4, scale: Nx2, sigma: float, far_plane: optional

B = 2  # batch size
N = 3  # number of quadmeshes per cuboid
H = 4  # image height
W = 5  # image width
C = 6  # feature channels
h = 3  # quadmesh feature height
w = 4  # quadmesh feature width

# normals: BxNx3
normals = torch.randn(B, N, 3)

# corners: BxNx4x3
corners = torch.randn(B, N, 4, 3)

# rays: HxWx3
rays = torch.randn(H, W, 3)

# quad_feat: BxNxCxhxw
quad_feat = torch.randn(B, N, C, h, w)

# quad2cam: BxNx4x4
quad2cam = torch.randn(B, N, 4, 4)

# scale: BxNx2
scale = torch.ones(B, N, 2) * 2.0

# sigma: float
sigma = 1e-05

# gamma: float
gamma = 0.0001

# near_plane: float
near_plane = 0.0

# far_plane: float or None
far_plane = 100.0

ARGS = {
    'normals': normals,
    'corners': corners,
    'rays': rays,
    'quad_feat': quad_feat,
    'quad2cam': quad2cam,
    'scale': scale,
    'sigma': sigma,
    'gamma': gamma,
    'near_plane': near_plane,
    'far_plane': far_plane,
}
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
project_faces Ran
ran on a synthesized input
autonomousvision/controllable_image_synthesis/controllable_gan/render/square.py
code served (permissive licence) · get_code("e69dfbbffbdc9e1d")
How it ran

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

import torch
import torch.nn as nn

torch.manual_seed(0)

# Image dimensions
im_height = 4
im_width = 5

# Number of quadrilateral faces
N = 2

# Number of feature channels
C = 3

# Create normal vectors for quadrilateral faces (Nx3)
n = torch.randn(N, 3)
n = n / (torch.norm(n, dim=1, keepdim=True) + 1e-6)

# Create center points of quadrilateral faces (Nx3)
p0 = torch.randn(N, 3)

# Create corners of quadrilateral faces (Nx4x3)
c = torch.randn(N, 4, 3)

# Create camera rays (HxWx3)
l = torch.randn(im_height, im_width, 3)
l = l / (torch.norm(l, dim=2, keepdim=True) + 1e-6)

# Create feature maps on the quadrilateral faces (NxCxHxW)
featmap = torch.randn(N, C, im_height, im_width)

# Create transformation matrix from quadmesh coordinate to cam coordinate (Nx4x4)
quad2cam = torch.eye(4).unsqueeze(0).expand(N, -1, -1).clone()
quad2cam[:, :3, :3] = torch.randn(N, 3, 3)
quad2cam[:, :3, 3] = torch.randn(N, 3)

# Create scale of quadmesh (N, 2)
scale = torch.ones(N, 2) * 2.0

# Sigma parameter for sharpness
sigma = 0.5

# Far plane parameter
far_plane = 100.0

ARGS = {
    'n': n,
    'p0': p0,
    'c': c,
    'l': l,
    'featmap': featmap,
    'quad2cam': quad2cam,
    'scale': scale,
    'sigma': sigma,
    'far_plane': far_plane,
}
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

Also present in other papers' code

Of 4 samples lifted from this paper's linked repositories, 4 carry a function identity and 0 do not. Of the 4 distinct functions with an identity, 1 are also present in other papers whose code we hold (1 with at least one paper dated earlier, 0 the same month, 1 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
actvn 6 other papers
1 dated earlier, 5 later
1810.00821 (2018-10)
Variational Discriminator Bottleneck: Improving Imitation Learning, Inverse RL, and GANs by Constraining Information Flow
2002.11810 (2020-02)
On Leveraging Pretrained GANs for Generation with Limited Data
2203.13572 (2022-03)
A Visual Navigation Perspective for Category-Level Object Pose Estimation
and 3 more not named here

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

In recent years, Generative Adversarial Networks have achieved impressive results in photorealistic image synthesis. This progress nurtures hopes that one day the classical rendering pipeline can be replaced by efficient models that are learned directly from images. However, current image synthesis models operate in the 2D domain where disentangling 3D properties such as camera viewpoint or object pose is challenging. Furthermore, they lack an interpretable and controllable representation. Our key hypothesis is that the image generation process should be modeled in 3D space as the physical world surrounding us is intrinsically three-dimensional. We define the new task of 3D controllable image synthesis and propose an approach for solving it by reasoning both in 3D space and in the 2D image domain. We demonstrate that our model is able to disentangle latent 3D factors of simple multi-object scenes in an unsupervised fashion from raw images. Compared to pure 2D baselines, it allows for synthesizing scenes that are consistent wrt. changes in viewpoint or object pose. We further evaluate various 3D representations in terms of their usefulness for this challenging task.

For agents

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

get_harvested_code_for_paper("1912.05237")
get_code_for_paper("1912.05237")
have("1912.05237")

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

curl https://syntology.ai/api/ran/1912.05237.json

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

[![Syntology run record](https://syntology.ai/api/ran/1912.05237.svg)](https://syntology.ai/paper/1912.05237)

Connect an agent — have() is free.