search

CASIA-IVA-Lab / FastSAM

Fast Segment Anything
GNU Affero General Public License v3.0
7.34k stars 682 forks source link

Do you have a plan to release FastSAM-n by training with YOLO-v8-n #165

Open sunwoo76 opened 1 year ago

sunwoo76 commented 1 year ago

Hello :)

I wonder that you have a plan to release the version of FastSAM-n with YOLOv8-n structure which is the smallest version of YOLOv8

Thank you.

zxDeepDiver commented 12 months ago

Currently, we think YOLOv8n is too small to perform well.

sunwoo76 commented 12 months ago

Currently, we think YOLOv8n is too small to perform well.

Thanks for replying the question :)

Additionally, is it possible to train the FastSAM with YOLOv8n with currently released code in this repository?

I'm trying to train the model but, the released code seems to be fitted to the COCO training. ( SA-1b's lable format is "json" but, the label is processed with "txt" format in the current training code.)

Thank you :)!

zxDeepDiver commented 11 months ago

Yes, convert the json to yolo's txt format.

sunwoo76 commented 11 months ago

Yes, convert the json to yolo's txt format.

In addition, Should the converted segmentation coordinates be normalized?

gahunghsu commented 8 months ago

Yes, convert the json to yolo's txt format.

How to convert the json to yolo's txt format?

sunwoo76 commented 8 months ago

Yes, convert the json to yolo's txt format.

How to convert the json to yolo's txt format?

convert compressed RLE format to normalized xy polygon format