search

med-air / 3DSAM-adapter

Holistic Adaptation of SAM from 2D to 3D for Promptable Medical Image Segmentation
170 stars 12 forks source link

LiT17 dataset source #1

Open cheliu-computation opened 1 year ago

cheliu-computation commented 1 year ago

Hi, authors, thanks for your impressive work on 3D SAM for medical image analysis.

I am downloading the LiT17 dataset from the original link provided in this repo, however the competition seems over and not allow for further downloading. Also I checked the processed data from your link. The processed data looks fine on /train folder, every volume has a seg file. But in /folder, there is only one file for each volume, from your dataset code, it seems the segment mask is combined with the raw volume,

I am not sure I understand correct, because the data format seems quite different with the original LiT17. Because I find the LiT17 format in this https://github.com/RL4M/PCRLv2/tree/finetune , and it shows every volume should pair with one seg file in train/test folder.

Looking forward to your reply!

peterant330 commented 1 year ago

Hi,

The LiTS 2017 released 131 volumes with labels for training and validation. For the test set, only the raw image is released. As a result, we did not use the test data in our paper. We split the original training data into our training/validation/test.

Actually, you can still download the data from the official website https://competitions.codalab.org/competitions/17094#participate-get-data, which only requires registration.

Best

Hi, authors, thanks for your impressive work on 3D SAM for medical image analysis.

I am downloading the LiT17 dataset from the original link provided in this repo, however the competition seems over and not allow for further downloading. Also I checked the processed data from your link. The processed data looks fine on /train folder, every volume has a seg file. But in /folder, there is only one file for each volume, from your dataset code, it seems the segment mask is combined with the raw volume,

I am not sure I understand correct, because the data format seems quite different with the original LiT17. Because I find the LiT17 format in this https://github.com/RL4M/PCRLv2/tree/finetune , and it shows every volume should pair with one seg file in train/test folder.

Looking forward to your reply!