Datasets:

---

ibrahimhamamci

/

CT-RATE

like 160

Note: I was able to download CT-RATE v1 BEFORE the v2 update (unable now as it exits) using snapshot_download.py from the example_download_script repo, just changed the script to have auto-restarts:

import time
from huggingface_hub import snapshot_download
from requests import ConnectionError

# Code taken from issue: https://github.com/huggingface/huggingface_hub/issues/1542

while True:
    try:
        snapshot_path =  snapshot_download(repo_id="ibrahimhamamci/CT-RATE", repo_type="dataset", local_dir="./dataset")
        break
    except Exception as e:
        print(f"ConnectionError: {e}. Retrying in 1s...")
        time.sleep(1)

same here, i tried any of the download script but got fetch 0 file (download_only_train, download_only_val, snapshot_download)

I'm not sure about the snapshot_download.py issue, but I downloaded v1 several months ago using the adapted version of the download_only_train_data.py, https://huggingface.co/datasets/ibrahimhamamci/CT-RATE/discussions/64 (farrell236 comment half way down). Just changing the directory to directory_name = f'dataset/{split}_fixed/' should work so you don't get the v1 download again. The v2 fixed version seems to be downloading properly for me currently this way. I see the script I used here: https://huggingface.co/datasets/ibrahimhamamci/CT-RATE/blob/main/download_scripts/download_dataset.py

What is the advantage of downloading using the sezginerr/example_download_script download scripts compared to git clone or huggingface-cli download?

updating this discussion - I updated the download_dataset.py script to download train_fixed and the segmentations (from TotalSegmentor only - ts_total, but can be used for all segs).
I put both scripts in a repo: https://github.com/yinon-gold/CT-RATE-download-scripts, planning to put a pull request to sezginerr/example_download_script.

The download_dataset.py is a small change from https://huggingface.co/datasets/ibrahimhamamci/CT-RATE/blob/main/download_scripts/download_dataset.py.

This is the script I used to download the TotalSegmentor segmentations, available in the repo as download_segmentations.py:

import shutil
import os
import pandas as pd
import sys
from huggingface_hub import hf_hub_download
from tqdm import tqdm

split = 'train_fixed'
batch_size = 100
start_at = 0

repo_id = 'ibrahimhamamci/CT-RATE'
directory_name = f'dataset/{split}/'
hf_token = os.getenv("HF_TOKEN")

data = pd.read_csv(f'{split.split("_")[0]}_labels.csv') # changed

for i in tqdm(range(start_at, len(data), batch_size)):

    data_batched = data[i:i+batch_size]

    for name in data_batched['VolumeName']:
        folder1 = name.split('_')[0]
        folder2 = name.split('_')[1]
        folder = folder1 + '_' + folder2
        folder3 = name.split('_')[2]
        subfolder = folder + '_' + folder3
        subfolder = directory_name + folder + '/' + subfolder
        ll = subfolder.split('/')  # added
        seg_folder = os.path.join(ll[0], 'ts_seg', 'ts_total', *ll[1:])  # added

        try:
            hf_hub_download(repo_id=repo_id,
                repo_type='dataset',
                token=hf_token,
                subfolder=seg_folder,
                filename=name,
                local_dir='<PATH_TO_CT-RATE>',
                resume_download=True,
            )
        except Exception as e:
            # append to file called errors.txt
            with open('errors.txt', 'a') as f:
                f.write(f"{e} :: {name=} {seg_folder=}\n")
            # print to stderr
            print(f"{e} :: {name=} {seg_folder=}", file=sys.stderr)
    try:
        # shutil.rmtree('.cache/huggingface/download/dataset/')  # note that a .cache folder is created and should be deleted
        continue
    except Exception as e:
        print(f"Error removing cache directory: {e}", file=sys.stderr)