search

zylye123 / CARes-UNet

CARes-UNet: Content-Aware Residual UNet for Lesion Segmentation of COVID-19 from Chest CT Images
MIT License
7 stars 3 forks source link

Release of checkpoints #2

Closed SimJeg closed 2 years ago

SimJeg commented 2 years ago

Hello,

Do you have any plan to release checkpoints of the models you trained ? I believe it would be very useful for the community and make life easier for benchmarks :) thank you for your work !

sysu19351158 commented 2 years ago

Hi,

Thank you for your suggestion. We havd updated the readme file and release the supervised checkpoint trained on COVID-SemiSeg Dataset for test demo. You can download it here “https://drive.google.com/file/d/1jv4OCiJGfWGRN3qdecJ4GXgH8uGdAkAw/view?usp=sharing”. Checkpoints trained on other datasets will be released soon.

Sorry to reply so late. Hopefully it will help you.

SimJeg commented 2 years ago

Hello,

I was able to run the model on new slices. Is there an optimal resolution and size to run the model ? From the code the size seems to be 256 pixels, for the resolution is it variable ? (e.g. 1.8mm / voxel). Also should we better crop images around the lungs ?

zylye123 commented 2 years ago

Hi,

Thank you for your question. If you want to use the checkpoint, the resolution and size should be strictly 256 * 256. If you prefer to use your own slices, you should at least resize or crop them first.

As for cropping around the lungs, we suggest that you could crop them like the Original Image from 4.3 Experiment Results in readme file, since the checkpoint was trained with these images. I assume that it would achieve better performance if you could finetune the pre-trained model first before you use it.

We are delighted to discuss about the related problems with you. Hopefully it would help you :)