SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2410.05057 · NeurIPS · 2024

SELECT: A Large-Scale Benchmark of Data Curation Strategies for Image Classification

Chinmay Hegde, Niv Cohen, Benjamin Feuer, Nyu, Jiawei Xu, Patrick Yubeaton, Govind Mittal, Imagenet++ Dataset, Utility Metrics

arXiv · PDF · Open in the Atlas

Code that ran

10 of 13 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; 1 honoured the contract we drafted, 1 violated it and 1 has no contract verdict; 7 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/2410.05057.json · badge: /api/ran/2410.05057.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
jimmyxu123/SELECT canonical (classifier label)
classifier basis: census
10 ran · 3 no recorded run
FunctionStatusWhere it lives
dedup_features Ran
ran; our contract draft was wrong, not the code
jimmyxu123/SELECT/ssl/utils.py
code served (permissive licence) · get_code("424fc184adb4a84c")
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. 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

torch.manual_seed(0)

# Create small sample tensors
# features: shape (batch=2, dim=4) - will have some duplicates to trigger dedup logic
features = torch.tensor([
    [1.0, 2.0, 3.0, 4.0],
    [1.0, 2.0, 3.0, 4.0],  # duplicate of first
], dtype=torch.float32)

# labels: shape (batch=2,) - can be any matching batch size
labels = torch.tensor([0, 1], dtype=torch.long)

ARGS = {
    "features": features,
    "labels": labels,
}
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_class_labels Ran
ran; our contract draft was wrong, not the code
jimmyxu123/SELECT/ssl/reader_hfds.py
code served (permissive licence) · get_code("1d74bc67f576b83a")
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 numpy as np
from collections import namedtuple

# Create a stand-in class for the label feature
class ClassLabel:
    def __init__(self, names):
        self.names = names
        self._str2int = {name: idx for idx, name in enumerate(names)}
    
    def str2int(self, name):
        return self._str2int[name]

# Create a stand-in class for features
class Features:
    def __init__(self):
        self._features = {
            'label': ClassLabel(['cat', 'dog', 'bird'])
        }
    
    def __contains__(self, key):
        return key in self._features
    
    def __getitem__(self, key):
        return self._features[key]

# Create a stand-in class for info
class DatasetInfo:
    def __init__(self):
        self.features = Features()

# Construct the info object
info = DatasetInfo()

ARGS = {
    'info': info,
    'label_key': 'label'
}
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
get_mean_and_std Ran
ran on a synthesized input
jimmyxu123/SELECT/vtab/getDataLoader.py
code served (permissive licence) · get_code("44aa6596c2101dce")
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 2, 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)

class SimpleDataset(torch.utils.data.Dataset):
    def __init__(self, num_samples=2):
        self.num_samples = num_samples
    
    def __len__(self):
        return self.num_samples
    
    def __getitem__(self, idx):
        # Return (input, target) tuple
        # Input: 3 channels, 4x4 spatial dimensions
        inputs = torch.randn(3, 4, 4)
        targets = torch.tensor(idx)
        return inputs, targets

dataset = SimpleDataset(num_samples=2)

ARGS = {
    'dataset': dataset
}
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 2, 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
load_class_map Ran
ran; our contract draft was wrong, not the code
jimmyxu123/SELECT/ssl/reader_hfds.py
code served (permissive licence) · get_code("2161fec9a47fe2fb")
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 os
import pickle
import tempfile

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

# Create a test .txt class map file
txt_map_path = os.path.join(temp_dir, 'classes.txt')
with open(txt_map_path, 'w') as f:
    f.write('cat\n')
    f.write('dog\n')
    f.write('bird\n')

# Create a test .pkl class map file
pkl_map_path = os.path.join(temp_dir, 'classes.pkl')
test_class_map = {'apple': 0, 'banana': 1, 'orange': 2}
with open(pkl_map_path, 'wb') as f:
    pickle.dump(test_class_map, f)

# Test with direct dict
dict_class_map = {'red': 0, 'green': 1, 'blue': 2}

ARGS = {
    'map_or_filename': dict_class_map,
    'root': temp_dir
}
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 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
log_and_continue Ran
ran, violated the contract we drafted
jimmyxu123/SELECT/analytical/quality-metrics.py
code served (permissive licence) · get_code("0a84b912bb131119")
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 bool, ndim 0, dtype family float, shape () 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)

# The function log_and_continue takes an exception as its only parameter
# It calls repr() on the exception and prints a message, then returns True
# So we need to pass a valid Exception instance

exn = ValueError("Test webdataset error")

ARGS = {
    "exn": exn
}
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 bool, ndim 0, dtype family float, shape () — shape () 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
regularize_label_count Ran
ran; our contract draft was wrong, not the code
jimmyxu123/SELECT/ssl/utils.py
code served (permissive licence) · get_code("e313a5d6604a698a")
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. 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

torch.manual_seed(0)

# Create small tensors for testing
# features: shape (num_samples, feature_dim)
# ret_labels: shape (num_samples,) with class indices
# label_freq: shape (num_classes,) with frequency counts per class

num_samples = 6
feature_dim = 4
num_classes = 3

features = torch.randn(num_samples, feature_dim)
ret_labels = torch.tensor([0, 0, 1, 1, 2, 2], dtype=torch.long)
label_freq = torch.tensor([2, 2, 2], dtype=torch.long)

ARGS = {
    "ret_labels": ret_labels,
    "features": features,
    "label_freq": label_freq,
    "thresh": 3
}
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
resample_features Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
jimmyxu123/SELECT/ssl/utils.py
code served (permissive licence) · get_code("a1db47a39780034c")
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 small tensor features (batch_size=6, feature_dim=4)
features = torch.randn(6, 4)

# Create labels with multiple classes (6 samples, 3 classes)
labels = torch.tensor([0, 0, 1, 1, 2, 2])

# Set return_size to trigger downsampling (less than total features)
return_size = 4

ARGS = {
    "features": features,
    "labels": labels,
    "return_size": return_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 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
test_loop Ran
ran, honoured the contract we drafted
jimmyxu123/SELECT/vtab/vtab_finetune.py
code served (permissive licence) · get_code("1b5c2979cb69dcb5")
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 float, ndim 0, dtype family float, shape () 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 as nn
from torch.utils.data import TensorDataset, DataLoader

torch.manual_seed(0)

# Create a small model
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

model = SimpleModel()

# Create a small dataset
X = torch.randn(2, 4)
y = torch.tensor([0, 1])
dataset = TensorDataset(X, y)

# Create a custom dataloader wrapper that has a .dl attribute
class DataLoaderWrapper:
    def __init__(self, dataloader):
        self.dl = dataloader
    
    def __iter__(self):
        return iter(self.dl)

dataloader_base = DataLoader(dataset, batch_size=2)
dataloader = DataLoaderWrapper(dataloader_base)

ARGS = {
    "dataloader": dataloader,
    "model": model
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
torch, tqdm — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type float, ndim 0, dtype family float, shape () — shape () 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
to_device Ran
ran; our contract draft was wrong, not the code
jimmyxu123/SELECT/vtab/getDataLoader.py
code served (permissive licence) · get_code("1c9147e50febcff4")
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)

# Test case 1: single tensor
single_tensor = torch.randn(2, 4)
device_cpu = 'cpu'

# Test case 2: list of tensors
tensor_list = [torch.randn(2, 4), torch.randn(3, 5)]

# Test case 3: tuple of tensors
tensor_tuple = (torch.randn(2, 4), torch.randn(3, 5))

# Test case 4: nested structure (list of tuples)
nested_data = [
    (torch.randn(2, 4), torch.randn(3, 5)),
    (torch.randn(2, 4), torch.randn(3, 5))
]

ARGS = {
    'data': single_tensor,
    'device': device_cpu
}
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
train_loop Ran
ran; our contract draft was wrong, not the code
jimmyxu123/SELECT/vtab/vtab_finetune.py
code served (permissive licence) · get_code("7b10e462ddbf8e58")
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 SimpleModel. 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.optim as optim

torch.manual_seed(0)

# Define a minimal model
class SimpleModel(nn.Module):
    def __init__(self):
        super().__init__()
        self.linear = nn.Linear(4, 3)
    
    def forward(self, x):
        return self.linear(x)

# Create small dataloader (list of tuples with (X, y))
dataloader = [
    (torch.randn(2, 4), torch.randn(2, 3)),
    (torch.randn(2, 4), torch.randn(2, 3)),
]

# Create model, loss function, optimizer, and scheduler
model = SimpleModel()
loss_fn = nn.MSELoss()
optimizer = optim.SGD(model.parameters(), lr=0.01)
scheduler = optim.lr_scheduler.StepLR(optimizer, step_size=10, gamma=0.1)
max_steps = 5

ARGS = {
    "dataloader": dataloader,
    "model": model,
    "loss_fn": loss_fn,
    "optimizer": optimizer,
    "scheduler": scheduler,
    "max_steps": max_steps,
}
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 SimpleModel — a SimpleModel output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a SimpleModel 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
easyrobust_eval Not yet run
unverified (harvested, no recorded run)
jimmyxu123/SELECT/base-ood/base_ood_eval.py
code served (permissive licence) · get_code("c7c1bff321bc2884")
How it ran

no invocation recorded.

get_data_loader Not yet run
unverified (harvested, no recorded run)
jimmyxu123/SELECT/vtab/getDataLoader.py
code served (permissive licence) · get_code("e0126d3f3d3d4c42")
How it ran

no invocation recorded.

knn_classifier Not yet run
unverified (harvested, no recorded run)
jimmyxu123/SELECT/ssl/eval_knn.py
code served (permissive licence) · get_code("6ecb6791e6a9ebfb")
How it ran

no invocation recorded.

Also present in other papers' code

Of 13 samples lifted from this paper's linked repositories, 13 carry a function identity and 0 do not. Of the 13 distinct functions with an identity, 2 are also present in other papers whose code we hold (2 with at least one paper dated earlier, 0 the same month, 1 later) and 11 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
knn_classifier 4 other papers
3 dated earlier, 1 later
2104.14294 (2021-04)
Emerging Properties in Self-Supervised Vision Transformers
2106.09681 (2021-06)
XCiT: Cross-Covariance Image Transformers
2109.13228 (2021-09)
PASS: An ImageNet replacement for self-supervised pretraining without humans
and 1 more not named here
to_device 2 other papers
2 dated earlier
2210.13043 (2022-10)
Data-IQ: Characterizing subgroups with heterogeneous outcomes in tabular data
2404.10824 (2024-04)
Decoupled Weight Decay for Any $p$ Norm

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

Data curation is the problem of how to collect and organize samples into a dataset that supports efficient learning. Despite the centrality of the task, little work has been devoted towards a large-scale, systematic comparison of various curation methods. In this work, we take steps towards a formal evaluation of data curation strategies and introduce SELECT , the first large-scale benchmark of curation strategies for image classification. In order to generate baseline methods for the SELECT benchmark, we create a new dataset, IMAGENET++ , which constitutes the largest superset of ImageNet-1K to date. Our dataset extends ImageNet with 5 new training-data shifts, each approximately the size of ImageNet-1K itself, and each assembled using a distinct curation strategy. We evaluate our data curation baselines in two ways: (i) using each training-data shift to train identical image classification models from scratch (ii) using it to inspect a fixed pretrained self-supervised representation. Our findings show interesting trends, particularly pertaining to recent methods for data curation such as synthetic data generation and lookup based on CLIP embeddings. We show that although these strategies are highly competitive for certain tasks, the curation strategy used to assemble the original ImageNet-1K dataset remains the gold standard. We anticipate that our benchmark can illuminate the path for new methods to further reduce the gap. We release our checkpoints, code, documentation, and a link to our dataset at https://github.com/jimmyxu123/SELECT.

For agents

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

get_harvested_code_for_paper("2410.05057")
get_code_for_paper("2410.05057")
have("2410.05057")

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

curl https://syntology.ai/api/ran/2410.05057.json

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

[![Syntology run record](https://syntology.ai/api/ran/2410.05057.svg)](https://syntology.ai/paper/2410.05057)

Connect an agent — have() is free.