Trainin result shows that val losses are 0

[darkstar1227]

• I am using Yolov8 and Yolov9 to do the comparation. Using the same datasets.
• Yolov8 has a normal result curve as Fig. 1.
• Yolov9 gave me that definitely wrong result curve as Fig. 2.
• I am using A100 to train it.
• Cuda 11.3, torch 1.11.0, pillow 9.5
• I have changed the pillow version because it shows a size error. I used the solution of yolov5.

[Youho99]

Check #132 Val Losses are not yet implemented for Yolov9 models Use val_dual.py instead

[Visiblemiracle]

try to reduce epochs number

[jasfa]

I trained yolov9-c with train_dual-py, where val_dual-py was used, and val/loss was still 0

[Kevin-Lab777]

@WongKinYiu 是用val_dual.py 应该也是0吧

[babakbch]

same problem here :(