SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2501.07171 · CVPR · 2025

BIOMEDICA: An Open Biomedical Image-Caption Archive, Dataset, and Vision-Language Models Derived from Scientific Literature

Jeffrey Gu, Yuhui Zhang, Xiaohan Wang, Robert Tibshirani, Liangyu Chen, Serena Yeung-Levy, Jeffrey Nirschl, James Burgess, Alejandro Lozano, Anita Austin, Min Woo, Ivan Lopez, and 4 more

arXiv · PDF · Open in the Atlas

Code that ran

16 of 20 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, 1 violated it and 15 have no contract verdict; 0 failures of our instrument; 4 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/2501.07171.json · badge: /api/ran/2501.07171.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
minwoosun/biomedica-etl canonical (classifier label)
classifier basis not recorded on the link
16 ran · 4 no recorded run
FunctionStatusWhere it lives
aggregate_publications_from_json Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/data_loaders/loader_utils.py
code served (permissive licence) · get_code("ff59295bb51c26d9")
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 tempfile
import json
from pathlib import Path

# Create a temporary directory structure with JSON files
temp_dir = tempfile.TemporaryDirectory()
temp_path = Path(temp_dir.name)
json_dir = temp_path / 'data' / 'json'
json_dir.mkdir(parents=True, exist_ok=True)

# Create sample JSON files with publication data
sample_data_1 = [
    {
        'figureset': [{'fig': 1}, {'fig': 2}],
        'title': 'Publication 1',
        'authors': 'Author A'
    },
    {
        'figureset': [],
        'title': 'Publication 2',
        'authors': 'Author B'
    }
]

sample_data_2 = [
    {
        'figureset': [{'fig': 3}],
        'title': 'Publication 3',
        'authors': 'Author C'
    }
]

# Write JSON files to disk
with open(json_dir / 'pub1.json', 'w') as f:
    json.dump(sample_data_1, f)

with open(json_dir / 'pub2.json', 'w') as f:
    json.dump(sample_data_2, f)

ARGS = {
    'directory_path': str(temp_path),
    'drop_empty_figure_sets': True,
    'n': None,
    'verbose': True
}
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 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
collect_tar_files Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/data_loaders/wd_loader.py
code served (permissive licence) · get_code("bcffe5ae51595c56")
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 os
import tempfile
import shutil

# Create a temporary directory structure with .tar files
temp_root = tempfile.mkdtemp()

# Create subset directories
subsets = ['commercial', 'noncommercial', 'other']
for subset in subsets:
    subset_dir = os.path.join(temp_root, subset)
    os.makedirs(subset_dir, exist_ok=True)
    
    # Create some dummy .tar files
    for i in range(2):
        tar_file = os.path.join(subset_dir, f'file{i}.tar')
        open(tar_file, 'w').close()

ARGS = {
    'root': temp_root,
    'subsets': ['commercial', 'noncommercial', 'other'],
    'random_seed': 42
}
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 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
collect_tar_files_from_labels Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/data_loaders/wd_loader.py
code served (permissive licence) · get_code("bbb9d8b536ae3d83")
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 os
import tempfile
import shutil

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

# Create subset directories
subsets = ['commercial', 'noncommercial', 'other']
for subset in subsets:
    subset_path = os.path.join(temp_dir, subset)
    os.makedirs(subset_path, exist_ok=True)
    
    # Create subdirectories within each subset
    for i in range(2):
        subdir_path = os.path.join(subset_path, f'subdir_{i}')
        os.makedirs(subdir_path, exist_ok=True)
        
        # Create some .tar files
        for j in range(2):
            tar_file_path = os.path.join(subdir_path, f'file_{j}.tar')
            with open(tar_file_path, 'w') as f:
                f.write('')

ARGS = {
    'root': temp_dir,
    'subsets': ['commercial', 'noncommercial', 'other'],
    'random_seed': 42
}
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 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
compare_csv_row_counts Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/download_batch.py
code served (permissive licence) · get_code("cb97752db69a8eba")
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 csv
import tempfile
import os

# Create temporary CSV files for testing
temp_dir = tempfile.mkdtemp()

# Create first CSV file with 5 rows (including header)
file1_path = os.path.join(temp_dir, "file1.csv")
with open(file1_path, 'w', newline='') as f:
    writer = csv.writer(f)
    writer.writerow(['col1', 'col2', 'col3'])
    writer.writerow(['a', 'b', 'c'])
    writer.writerow(['d', 'e', 'f'])
    writer.writerow(['g', 'h', 'i'])
    writer.writerow(['j', 'k', 'l'])

# Create second CSV file with 5 rows (including header) - matching count
file2_path = os.path.join(temp_dir, "file2.csv")
with open(file2_path, 'w', newline='') as f:
    writer = csv.writer(f)
    writer.writerow(['x', 'y', 'z'])
    writer.writerow(['1', '2', '3'])
    writer.writerow(['4', '5', '6'])
    writer.writerow(['7', '8', '9'])
    writer.writerow(['10', '11', '12'])

ARGS = {
    'file1_path': file1_path,
    'file2_path': file2_path
}
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 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
create_log_file_path Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/build_json.py
code served (permissive licence) · get_code("d83d88e5f98e49d5")
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 str. 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

# Set up the arguments for create_log_file_path
directory = "/path/to/batch_data/sample_batch_001"
batch_index = 5
log_type = "error"
log_dir = "/tmp/logs/"

ARGS = {
    "directory": directory,
    "batch_index": batch_index,
    "log_type": log_type,
    "log_dir": log_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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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
extract_features_from_dataloader Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/feature_extractors/extract_features.py
code served (permissive licence) · get_code("ac1c2471664def4c")
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
import torch.nn as nn
import numpy as np
import tempfile
import os

torch.manual_seed(0)

# Create a minimal model that satisfies the function's usage
class DummyFeatureExtractor(nn.Module):
    def __init__(self, input_dim=3, output_dim=8):
        super().__init__()
        self.linear = nn.Linear(input_dim, output_dim)
    
    def forward(self, x):
        # x shape: (batch, channels, height, width) -> flatten to (batch, channels*height*width)
        batch_size = x.shape[0]
        x_flat = x.view(batch_size, -1)
        return self.linear(x_flat)

# Create a minimal dataloader
# Each sample is (image, label, key)
def create_dummy_dataloader(num_batches=5, batch_size=2):
    batches = []
    for b in range(num_batches):
        images = torch.randn(batch_size, 3, 4, 4)  # (batch, channels, height, width)
        labels = torch.randint(0, 10, (batch_size,))
        keys = [f"key_{b}_{i}" for i in range(batch_size)]
        batches.append((images, labels, keys))
    return batches

# Build arguments
model = DummyFeatureExtractor(input_dim=3*4*4, output_dim=8)
dataloader = create_dummy_dataloader(num_batches=5, batch_size=2)
save_dir = tempfile.mkdtemp()

ARGS = {
    'dataloader': dataloader,
    'model': model,
    'device': 'cpu',
    'max_images': 3,
    'store_images': 2,
    'save': save_dir,
    'save_every': 2,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
numpy, torch, tqdm — 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
extract_image_caption_pairs Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/data_loaders/loader_utils.py
code served (permissive licence) · get_code("e0bc84a5913a68cd")
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 numpy as np
from typing import List, Dict, Any, Tuple

# Construct sample publications data
publications = [
    {
        'accession_id': 'ACC001',
        'figureset': [
            {
                'image_path': '/path/to/image1.jpg',
                'caption': 'This is the first image caption',
                'image_id': 'IMG001'
            },
            {
                'image_path': '/path/to/image2.jpg',
                'caption': 'This is the second image caption',
                'image_id': 'IMG002'
            }
        ]
    },
    {
        'accession_id': 'ACC002',
        'figureset': [
            {
                'image_path': '/path/to/image3.jpg',
                'caption': 'Third image caption here',
                'image_id': 'IMG003'
            }
        ]
    },
    {
        'accession_id': 'ACC003',
        'figureset': [
            {
                'image_path': '/path/to/image4.jpg',
                'caption': 'Fourth image caption',
                'image_id': 'IMG004'
            },
            {
                'image_path': '/path/to/image5.jpg',
                'image_id': 'IMG005'
            }
        ]
    }
]

ARGS = {
    'publications': publications
}
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 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
filter_no_caption_or_no_image Ran
ran, violated the contract we drafted
minwoosun/biomedica-etl/src/pmc_oa/data_loaders/wd_loader.py
code served (permissive licence) · get_code("38a36c2da131f507")
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

# Define a sample dictionary with both caption and image
sample_valid = {
    'txt': 'A cat sitting on a mat',
    'jpg': b'fake_image_data'
}

# Define ARGS mapping the parameter to a value
ARGS = {
    'sample': sample_valid
}
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
get_image_identifier Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/caption.py
code served (permissive licence) · get_code("42b8eab2675d547e")
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 os

# The function takes a list of file paths as strings
# It extracts file names without extensions from each path
# Let's create a list of image file paths

list_of_path_to_image = [
    "/home/user/images/photo1.jpg",
    "/home/user/images/photo2.png",
    "/home/user/images/photo3.jpeg",
    "relative/path/image.gif",
    "another_image.bmp"
]

ARGS = {
    "list_of_path_to_image": list_of_path_to_image
}
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 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
get_image_paths Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/caption.py
code served (permissive licence) · get_code("7a900ce20281ac4b")
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 os
import tempfile
import shutil

# Create a temporary directory with some test image files
temp_dir = tempfile.mkdtemp()

# Create some .jpg files in the temp directory
jpg_file_1 = os.path.join(temp_dir, "image1.jpg")
jpg_file_2 = os.path.join(temp_dir, "image2.jpg")
non_jpg_file = os.path.join(temp_dir, "document.txt")

# Write dummy content to the files
with open(jpg_file_1, 'w') as f:
    f.write("dummy image data")
with open(jpg_file_2, 'w') as f:
    f.write("dummy image data")
with open(non_jpg_file, 'w') as f:
    f.write("dummy text data")

ARGS = {
    "path_to_pmcid": 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 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
get_kmeans Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/label_utils/models.py
code served (permissive licence) · get_code("544ccd5b57784198")
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 ndarray, ndim 1, dtype family int, shape (10,) 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
import tempfile
import os

np.random.seed(0)

# Create small feature array: 10 samples, 5 features
features = np.random.rand(10, 5).astype(np.float32)

# Create a temporary directory for potential file operations
temp_dir = tempfile.gettempdir()
save_path = os.path.join(temp_dir, "kmeans_test_model.joblib")

ARGS = {
    "features": features,
    "num_clusters": 3,
    "save_to": save_path,
    "from_file": False,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
joblib, numpy, sklearn — 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 int, shape (10,) — shape (10,) 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
get_main_batch_index Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/download_batch.py
code served (permissive licence) · get_code("2e665f86d4a45067")
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 str. 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 get_main_batch_index takes a sub_batch_id string and extracts
# the main batch index by splitting on underscore and returning the first part.
# It handles both cases: with underscore (returns first part) and without (returns original).

# Test case 1: sub_batch_id with underscore
sub_batch_id_with_underscore = "batch_001_subset"

# Test case 2: sub_batch_id without underscore
sub_batch_id_without_underscore = "batch_001"

# We'll use the first one as the primary argument
ARGS = {
    "sub_batch_id": "batch_001_subset"
}
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 str — a str output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a str 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_next_index Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/build_json.py
code served (permissive licence) · get_code("180bb7762557abf2")
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 int, ndim 0, dtype family int, 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 os
import re
import tempfile

# Create a temporary directory with some test JSON files
temp_dir = tempfile.mkdtemp()

# Create some test files matching the expected pattern
test_filelist_name = "test_filelist"
for i in range(3):
    filepath = os.path.join(temp_dir, f"{test_filelist_name}_{i}.json")
    with open(filepath, 'w') as f:
        f.write('{}')

ARGS = {
    "json_dir": temp_dir,
    "filelist_name": test_filelist_name,
}
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 int, ndim 0, dtype family int, 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
get_pca Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/label_utils/models.py
code served (permissive licence) · get_code("b1a84bc8eee5b294")
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 ndarray, ndim 2, dtype family float, shape (100, 30) 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
import tempfile
import os

# Create sample features with shape (100, 50) - 100 samples, 50 features
features = np.random.rand(100, 50)

# Create temporary file paths for save_to and from_file
temp_dir = tempfile.gettempdir()
save_path = os.path.join(temp_dir, "pca_model_test.joblib")
from_file_path = None  # We'll set this to None for the initial run, or to save_path for a second run

ARGS = {
    "features": features,
    "from_file": from_file_path,
    "save_to": save_path,
    "n_components": 30
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
joblib, numpy, sklearn — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (100, 30) — shape (100, 30) 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
load_and_process_media_filelist_csv Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/download.py
code served (permissive licence) · get_code("cca1e5f8e9f7b0e0")
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 DataFrame. 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 pandas as pd
import tempfile
import os

# Create a temporary CSV file with the required structure
temp_dir = tempfile.mkdtemp()
csv_path = os.path.join(temp_dir, "media_filelist.csv")

# Create a sample CSV with File column
csv_data = pd.DataFrame({
    'File': [
        'path/to/media1.mp4',
        'path/to/media2.wav',
        'another/path/media3.jpg'
    ]
})
csv_data.to_csv(csv_path, index=False)

ARGS = {
    'path_to_filelist': csv_path
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1, through the batch API
Imports
pandas — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type DataFrame — a DataFrame output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a DataFrame output the recorded ndim 2 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
read_json_file Ran
ran on a synthesized input
minwoosun/biomedica-etl/src/pmc_oa/data_loaders/loader_utils.py
code served (permissive licence) · get_code("97c7f58efcfba433")
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 json
import tempfile
from pathlib import Path

# Create a temporary JSON file with valid data
temp_file = tempfile.NamedTemporaryFile(mode='w', suffix='.json', delete=False)
test_data = [
    {"id": 1, "name": "Alice", "value": 42},
    {"id": 2, "name": "Bob", "value": 137}
]
json.dump(test_data, temp_file)
temp_file.close()

ARGS = {
    "file_path": Path(temp_file.name)
}
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 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
download_media Not yet run
unverified (harvested, no recorded run)
minwoosun/biomedica-etl/src/pmc_oa/download.py
code served (permissive licence) · get_code("7a342b3fd8f33439")
How it ran

no invocation recorded.

load_and_concatenate_subset Not yet run
unverified (harvested, no recorded run)
minwoosun/biomedica-etl/src/pmc_oa/data_loaders/dino_features_loader.py
code served (permissive licence) · get_code("f2f23afd2de6f9dc")
How it ran

no invocation recorded.

load_and_concatenate_subsets Not yet run
unverified (harvested, no recorded run)
minwoosun/biomedica-etl/src/pmc_oa/data_loaders/dino_features_loader.py
code served (permissive licence) · get_code("ae69a8e47ba31b5d")
How it ran

no invocation recorded.

reconnect_ftp Not yet run
unverified (harvested, no recorded run)
minwoosun/biomedica-etl/src/pmc_oa/download.py
code served (permissive licence) · get_code("1a9cc446f390c491")
How it ran

no invocation recorded.

Also present in other papers' code

Of 20 samples lifted from this paper's linked repositories, 20 carry a function identity and 0 do not. Of the 20 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, 0 later) and 19 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
filter_no_caption_or_no_image 1 other paper
1 dated earlier
2407.11867 (2024-07)
Targeted Unlearning with Single Layer Unlearning Gradient

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

The development of vision-language models (VLMs) is driven by large-scale and diverse multi-modal datasets. However, progress toward generalist biomedical VLMs is limited by the lack of annotated, publicly accessible datasets across biology and medicine. Existing efforts are limited to narrow domains, missing the full diversity of biomedical knowledge encoded in scientific literature. To address this gap, we introduce BIOMEDICA: a scalable, open-source framework to extract, annotate, and serialize the entirety of the PubMed Central Open Access subset into an easy-to-use, publicly accessible dataset. Our framework produces a comprehensive archive with over 24 million unique image-text pairs from over 6 million articles. Metadata and expert-guided annotations are additionally provided. We demonstrate the utility and accessibility of our resource by releasing BMC-CLIP, a suite of CLIP-style models continuously pre-trained on BIOMEDICA dataset via streaming (eliminating the need to download 27 TB of data locally). On average, our models achieve state-of-theart performance across 40 tasks -spanning pathology, radiology, ophthalmology, dermatology, surgery, molecular biology, parasitology, and cell biology -excelling in zero-shot classification with 6.56% average improvement (as high as 29.8% and 17.5% in dermatology and ophthalmology, respectively) and stronger image-text retrieval while using 10x less compute. To foster reproducibility and collaboration, we release our codebase 1 , 2 and dataset 3 to the broader research community.

For agents

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

get_harvested_code_for_paper("2501.07171")
get_code_for_paper("2501.07171")
have("2501.07171")

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

curl https://syntology.ai/api/ran/2501.07171.json

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

[![Syntology run record](https://syntology.ai/api/ran/2501.07171.svg)](https://syntology.ai/paper/2501.07171)

Connect an agent — have() is free.