askerlee/segtran

Are checkpoints of this model available?

MengzhangLI opened this issue · 9 comments

MengzhangLI commented

Hi, thanks for your excellent work!

I want to re-implement this model, will you provide your trained models whose results achieve SOTA from your paper? Thus we can only use your model doing inference.

Especially model on REFUGE and BraTS, segtran is very extraordinary.

Best,

askerlee commented

Sure I'll be happy to share the checkpoints. Lemme check whether they are still on my harddisk. If not, I'll do a retraining.

RylanSchaeffer commented

@askerlee any updates? :D

askerlee commented

@RylanSchaeffer I uploaded a checkpoint for REFUGE to baidu netdisk:
URL: https://pan.baidu.com/s/1TmV27ziwrULGa0KFA2ohuQ
code: qsll

The problem is it's too big (around 1G). I haven't checked why it's so big.
It achieves 0.959/0.870 on REFUGE leaderboard.

wshi8 commented

@askerlee Thank you for loading the model, would you please also upload the trained model checkpoint for the BraTS? I am looking into using the 3D model for the BraTS dataset

askerlee commented

I've deleted the checkpoint but I'll retrain it today. Probably will be able to upload it tomorrow.

wshi8 commented

Thank you @askerlee

askerlee commented

@wshi8 The checkpoints on BraTS 2019 and REFUGE 2020 have been uploaded to:
https://pan.baidu.com/s/1nbBrPJKK1NtOe848cApS8w
code: h1vi

On BraTS 2019 validation:
0.729/0.896/0.832, Avg. 0.819
On REFUGE 2020 validation:
0.870/0.959, Avg. 0.915

wshi8 commented

Thanks a lot! what is the 3 value number corresponding to? is this "Accuracy" for 3 validation runs?
0.729/0.896/0.832, Avg. 0.819

askerlee commented

@wshi8 They are the dice scores of ET (enhancing tumor), WT (whole tumor) and TC (tumor core), just as the metrics reported in my ijcai paper.