2 questions

[SylvainArd]

Hello,
is yolov9 able to do instance segmentation, I doubt of it !
the format of TXT for instance segmentation is :

• a file per image with same name but TXT extension
• a line per polygon with first number the class and others the X and Y coordinates of each point of the polygon, the Xs divided by the width of the image, the Ys divided by the height of the image
  In other website I see that after the class number and before the polygon coordinates we must put the bbox coordinates normalized is it true ?

isn't it ?
Thank you
Best regards

[SylvainArd]

I found the solution

[Youho99]

I found the solution

Can you give the solution, for those who might have the same question

And you can close the issue

[SylvainArd]

yes ! for the prediction I made this change :
pred, proto = model(im, augment=augment, visualize=visualize)[:2]
proto = proto[2]#added by SA
unless I had a bug "list has not attribute shape"
and for the TXT file it is :

a file per image with same name but TXT extension
a line per polygon with first number the class and others the X and Y coordinates of each point of the polygon, the Xs divided by the width of the image, the Ys divided by the height of the image
In other website I see that after the class number and before the polygon coordinates we must put the bbox coordinates normalized is it true ? >> FALSE

I made a train.bat with :
call conda activate yolov9
python segment/train.py --workers 8 --device 0 --batch 6 --data coco.yaml --img 640 --cfg models/segment/gelan-c-seg.yaml --weights '' --name gelan-c-seg --hyp hyp.scratch-high.yaml --no-overlap --epochs 300000 --close-mosaic 10
pause

and a predict.bat with :
call conda activate yolov9
python segment/predict.py --weights "C:\Users\Sylvain\Desktop\yolov9-main\best.pt" --source "data/images" --conf-thres 0.6 --max-det 1000
pause

I putted all my TXT and JPG files in yolov9-main/datasets/yolo_polygons in the same directory
finally I changed the data/coco.yaml according to my dataset
For predicting I put the best.pt from yolov9-main\runs\train-seg\gelan-c-seg12\weights to root yolov9 directory, NB change 12 to your training try