SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2303.09165 · CVPR · 2023

A New Benchmark: On the Utility of Synthetic Data with Blender for Bare Supervised Learning and Downstream Domain Adaptation

Hui Tang, Kui Jia

arXiv · PDF · Open in the Atlas

Code that ran

9 of 12 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; 2 honoured the contract we drafted, 2 violated it and 4 have no contract verdict; 1 failure 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/2303.09165.json · badge: /api/ran/2303.09165.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
huitangtang/on_the_utility_of_synthetic_data canonical (classifier label)
classifier basis not recorded on the link
9 ran · 3 no recorded run
FunctionStatusWhere it lives
FeedForward Ran
ran on a synthesized input
huitangtang/on_the_utility_of_synthetic_data/supervised_learning/models/mlp_mixer.py
code served (permissive licence) · get_code("3774eb4ab31e4b07")
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 Sequential. 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)

# Analyze the function:
# - dim: int, input/output dimension
# - expansion_dim: int, hidden dimension
# - dropout: float, dropout rate (default 0.0)
# - dense: callable that creates a linear layer (default nn.Linear)

# The function creates a sequential model:
# dense(dim, expansion_dim) -> GELU -> Dropout -> dense(expansion_dim, dim) -> Dropout
# So dim and expansion_dim should be distinct integers
# dense should be a callable that takes (in_features, out_features) and returns a layer

ARGS = {
    "dim": 4,
    "expansion_dim": 8,
    "dropout": 0.1,
    "dense": nn.Linear,
}
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 Sequential — a Sequential output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a Sequential output the recorded ndim 1 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
accuracy Ran
ran on a synthesized input
huitangtang/on_the_utility_of_synthetic_data/domain_adaptation/DANN/trainer.py
code served (permissive licence) · get_code("bf74514ea371233a")
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.

import torch

torch.manual_seed(0)

# Create small tensors for the accuracy function
batch_size = 2
num_classes = 4

# output: logits/scores tensor of shape (batch_size, num_classes)
output = torch.randn(batch_size, num_classes)

# target: ground truth class indices of shape (batch_size,)
target = torch.tensor([1, 3])

# topk: tuple of k values to compute accuracy for
topk = (1, 2)

ARGS = {
    "output": output,
    "target": target,
    "topk": topk,
}
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 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
bool_flag Ran
ran, violated the contract we drafted
huitangtang/on_the_utility_of_synthetic_data/supervised_learning/utils.py
pointer only (licence: MIT); the stored copy get_code serves, from sthalles/MaSSL, is pointer only (licence: NONE) · get_code("c19231378e41dc5a")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from sthalles/MaSSL, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

clip_gradients Ran
ran, honoured the contract we drafted
huitangtang/on_the_utility_of_synthetic_data/supervised_learning/utils.py
code served (permissive licence) · get_code("2b5489053b222e00")
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.

import torch
import torch.nn as nn

torch.manual_seed(0)

class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear1 = nn.Linear(4, 3)
        self.linear2 = nn.Linear(3, 2)
    
    def forward(self, x):
        return self.linear2(self.linear1(x))

model = SimpleModel()

# Create a simple forward pass and backward pass to populate gradients
x = torch.randn(2, 4)
y = torch.randn(2, 2)
output = model(x)
loss = nn.MSELoss()(output, y)
loss.backward()

clip = 1.0

ARGS = {
    "model": model,
    "clip": clip,
}
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 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
cosine_scheduler Ran
ran, honoured the contract we drafted
huitangtang/on_the_utility_of_synthetic_data/supervised_learning/utils.py
code served (permissive licence) · get_code("361a6b24f11fc50a")
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) and returned ndarray, ndim 1, dtype family float, shape (500,) 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

ARGS = {
    "base_value": 0.1,
    "final_value": 0.001,
    "epochs": 10,
    "niter_per_ep": 50,
    "warmup_epochs": 2,
    "start_warmup_value": 0.0,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 (500,) — shape (500,) 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
drop_path Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
huitangtang/on_the_utility_of_synthetic_data/supervised_learning/models/vision_transformer.py
pointer only (licence: MIT); the stored copy get_code serves, from gladzhang/Xformer, is pointer only (licence: NONE) · get_code("55120f2026b56aa2")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from gladzhang/Xformer, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

evaluate Ran
ran on a synthesized input
huitangtang/on_the_utility_of_synthetic_data/domain_adaptation/DisClusterDA/trainer.py
code served (permissive licence) · get_code("662f238409af534f")
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. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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 os
import tempfile

torch.manual_seed(0)

# Create a temporary directory for logs
temp_dir = tempfile.mkdtemp()

# Build a minimal model that returns a list with output as last element
class SimpleModel(nn.Module):
    def __init__(self, num_classes=4):
        super(SimpleModel, self).__init__()
        self.fc = nn.Linear(5, num_classes)
    
    def forward(self, x):
        # Function expects output = model(input_var)[-1]
        # So we return a list with the actual output as the last element
        output = self.fc(x)
        return [output, output]  # Return list, last element is the logits

# Create a simple criterion
criterion = nn.CrossEntropyLoss()

# Create a validation loader (list of batches with input and target)
# Each batch is (input_tensor, target_tensor)
val_loader = []
batch_size = 2
input_dim = 5
num_classes = 4

for _ in range(3):
    input_batch = torch.randn(batch_size, input_dim)
    target_batch = torch.randint(0, num_classes, (batch_size,))
    val_loader.append((input_batch, target_batch))

# Create model
model = SimpleModel(num_classes=num_classes)

# Create args dict with all required attributes
class Args:
    def __init__(self):
        self.num_classes = 4
        self.print_freq = 2
        self.log = temp_dir

args = Args()

# Build ARGS dict
ARGS = {
    'val_loader': val_loader,
    'model': model,
    'criterion': criterion,
    'epoch': 1,
    'args': args
}
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 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
pair Ran
ran, violated the contract we drafted
huitangtang/on_the_utility_of_synthetic_data/supervised_learning/models/vit.py
code served (permissive licence) · get_code("6ba8cee9f5daea41")
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) and returned tuple, ndim 1, dtype family int; 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)

# The function 'pair' takes a single parameter 't' and returns it as a tuple if it isn't already one
# We need to test both cases: when t is a tuple and when it's not

# Case 1: t is not a tuple (e.g., an integer, string, tensor, etc.)
t_not_tuple = 5

# Case 2: t is already a tuple
t_is_tuple = (3, 7)

# We'll use the non-tuple case as the primary test
ARGS = {
    "t": t_not_tuple
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
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 1, dtype family int — 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
train Ran
ran on a synthesized input
huitangtang/on_the_utility_of_synthetic_data/domain_adaptation/DANN/trainer.py
code served (permissive licence) · get_code("a63b80aa6e20d75f")
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 1, dtype family float; no output shape is stated. The image it ran in was not recorded. The code mentions cuda, and the harness aliases cuda to the CPU for every sample; whether this run needed that is 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
import torch.nn as nn
import math
import time

torch.manual_seed(0)

# Create a simple model that returns two outputs
class SimpleModel(nn.Module):
    def __init__(self):
        super(SimpleModel, self).__init__()
        self.fc1 = nn.Linear(4, 8)
        self.fc_lp = nn.Linear(8, 3)  # label predictor output
        self.fc_dc = nn.Linear(8, 2)  # domain classifier output
    
    def forward(self, x, lam):
        x = self.fc1(x)
        output_lp = self.fc_lp(x)
        output_dc = self.fc_dc(x)
        return output_lp, output_dc

# Create data loaders (list of batches)
def create_data_loader(batch_size=2, num_batches=1, input_dim=4, num_classes=3):
    loader = []
    for i in range(num_batches):
        x = torch.randn(batch_size, input_dim)
        y = torch.randint(0, num_classes, (batch_size,))
        loader.append((x, y))
    return loader

# Simple AverageMeter-like class for tracking metrics
class AverageMeter:
    def __init__(self):
        self.val = 0
        self.avg = 0
        self.sum = 0
        self.count = 0
    
    def update(self, val, n=1):
        self.val = val
        self.sum += val * n
        self.count += n
        self.avg = self.sum / self.count

# Create args object
class Args:
    def __init__(self):
        self.epochs = 10
        self.print_freq = 1
        self.log = '/tmp'

# Create directory for logging
import os
os.makedirs('/tmp', exist_ok=True)

# Setup all arguments
model = SimpleModel()
criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01)

source_train_loader = create_data_loader(batch_size=2, num_batches=2, input_dim=4, num_classes=3)
target_train_loader = create_data_loader(batch_size=2, num_batches=2, input_dim=4, num_classes=3)

source_train_loader_batch = enumerate(source_train_loader)
target_train_loader_batch = enumerate(target_train_loader)

train_records = {
    'data_time': AverageMeter(),
    'batch_time': AverageMeter(),
    'losses_overall': AverageMeter(),
    'losses_lp': AverageMeter(),
    'losses_dc': AverageMeter(),
    'top1_lp': AverageMeter(),
    'top1_dc': AverageMeter(),
}

args = Args()

ARGS = {
    'source_train_loader': source_train_loader,
    'source_train_loader_batch': source_train_loader_batch,
    'target_train_loader': target_train_loader,
    'target_train_loader_batch': target_train_loader_batch,
    'model': model,
    'criterion': criterion,
    'optimizer': optimizer,
    'itern': 0,
    'current_epoch': 1,
    'train_records': train_records,
    'args': args,
}
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 1, 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
accuracy Not yet run
unverified (harvested, no recorded run)
huitangtang/on_the_utility_of_synthetic_data/domain_adaptation/DisClusterDA/trainer.py
pointer only (licence: MIT); the stored copy get_code serves, from kkonstantinidis/ByzShield, is pointer only (licence: NONE) · get_code("4d3faaaa1c706b0b")
How it ran

The body of this sample is withheld: this paper's link allows the body inline (MIT), but get_code serves the stored copy, first harvested from kkonstantinidis/ByzShield, and GitHub detected no licence in that repository when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

train_compute_class_mean Not yet run
unverified (harvested, no recorded run)
huitangtang/on_the_utility_of_synthetic_data/domain_adaptation/DisClusterDA/trainer.py
code served (permissive licence) · get_code("a977faa9603e6bd1")
How it ran

no invocation recorded.

validate Not yet run
unverified (harvested, no recorded run)
huitangtang/on_the_utility_of_synthetic_data/domain_adaptation/DANN/trainer.py
code served (permissive licence) · get_code("9a04ba7ad97ceffd")
How it ran

no invocation recorded.

Also present in other papers' code

Of 12 samples lifted from this paper's linked repositories, 12 carry a function identity and 0 do not. Of the 12 distinct functions with an identity, 6 are also present in other papers whose code we hold (6 with at least one paper dated earlier, 5 the same month, 6 later) and 6 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
drop_path 224 other papers
75 dated earlier, 10 the same month, 132 later, 6 same year, order not determinable, 1 undated · 38 only as an assembly brick
2003.12039 (2020-03)
RAFT: Recurrent All-Pairs Field Transforms for Optical Flow
2007.09384 (2020-07)
Multi-Scale Positive Sample Refinement for Few-Shot Object Detection
2010.11929 (2020-10)
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
and 221 more not named here
pair 90 other papers
27 dated earlier, 1 the same month, 59 later, 2 same year, order not determinable, 1 undated · 8 only as an assembly brick
2004.13649 (2020-04)
Image Augmentation Is All You Need: Regularizing Deep Reinforcement Learning from Pixels
2010.11929 (2020-10)
An Image is Worth 16x16 Words: Transformers for Image Recognition at Scale
2103.15691 (2021-03)
ViViT: A Video Vision Transformer
and 87 more not named here
cosine_scheduler 37 other papers
13 dated earlier, 2 the same month, 22 later
2005.10242 (2020-05)
Understanding Contrastive Representation Learning through Alignment and Uniformity on the Hypersphere
2103.15375 (2021-03)
AlignMixup: Improving Representations By Interpolating Aligned Features
2106.09681 (2021-06)
XCiT: Cross-Covariance Image Transformers
and 34 more not named here
bool_flag 31 other papers
9 dated earlier, 3 the same month, 19 later
2106.09681 (2021-06)
XCiT: Cross-Covariance Image Transformers
2109.13228 (2021-09)
PASS: An ImageNet replacement for self-supervised pretraining without humans
2111.12958 (2021-11)
Self-Distilled Self-Supervised Representation Learning
and 28 more not named here
clip_gradients 28 other papers
9 dated earlier, 2 the same month, 17 later
2106.09681 (2021-06)
XCiT: Cross-Covariance Image Transformers
2109.13228 (2021-09)
PASS: An ImageNet replacement for self-supervised pretraining without humans
2111.12958 (2021-11)
Self-Distilled Self-Supervised Representation Learning
and 25 more not named here
accuracy 23 other papers
21 dated earlier, 2 later
1603.07285 (2016-03)
A guide to convolution arithmetic for deep learning
1608.07017 (2016-08)
Ambient Sound Provides Supervision for Visual Learning
1608.08710 (2016-08)
Pruning Filters for Efficient ConvNets
and 20 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

Deep learning in computer vision has achieved great success with the price of large-scale labeled training data. However, exhaustive data annotation is impracticable for each task of all domains of interest, due to high labor costs and unguaranteed labeling accuracy. Besides, the uncontrollable data collection process produces non-IID training and test data, where undesired duplication may exist. All these nuisances may hinder the verification of typical theories and exposure to new findings. To circumvent them, an alternative is to generate synthetic data via 3D rendering with domain randomization. We in this work push forward along this line by doing profound and extensive research on bare supervised learning and downstream domain adaptation. Specifically, under the well-controlled, IID data setting enabled by 3D rendering, we systematically verify the typical, important learning insights, e.g., shortcut learning, and discover the new laws of various data regimes and network architectures in generalization. We further investigate the effect of image formation factors on generalization, e.g., object scale, material texture, illumination, camera viewpoint, and background in a 3D scene. Moreover, we use the simulation-to-reality adaptation as a downstream task for comparing the transferability between synthetic and real data when used for pre-training, which demonstrates that synthetic data pre-training is also promising to improve real test results. Lastly, to promote future research, we develop a new large-scale synthetic-to-real benchmark for image classification, termed S2RDA, which provides more significant challenges for transfer from simulation to reality.

For agents

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

get_harvested_code_for_paper("2303.09165")
get_code_for_paper("2303.09165")
have("2303.09165")

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

curl https://syntology.ai/api/ran/2303.09165.json

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

[![Syntology run record](https://syntology.ai/api/ran/2303.09165.svg)](https://syntology.ai/paper/2303.09165)

Connect an agent — have() is free.