Open Haoyanlong opened 4 years ago
The loss functions are not normalized. e.t. are not divided bt (bhw*c)
@Haoyanlong As @xiong224 mentioned, we didn't normalize the loss function during training. Therefore, the value of loss is large, and when it finally converges, the value will be between 1e4 and 1e5
@Haoyanlong As @xiong224 mentioned, we didn't normalize the loss function during training. Therefore, the value of loss is large, and when it finally converges, the value will be between 1e4 and 1e5
@vickyFox @xiong224 OK, I see!Thank you very much!
@vickyFox ,Thank you for your work!I have trained the model in my dataset?But the loss is very big(10e5) as follow!Could you give me some advice?