Model not learning #10
Comments
|
|
|
As shown above, it looks like the classifier (l_cla) is suffering from overfitting. That is, comparing EVAL to tr-gen values |
|
hi ,I got that same problem even I trained for 10000 epochs after 4000 epochs nothing much change . Can i have some tricks |
|
The pretrained model is referenced in this issue: #5 |
hello Bird9000: Sorry for the late reply. My problem was that I was new to GAN training, and so, I focused on metrics. As training progresses, what you have to do is observe the intermediate results in imgs/ folder. Even though the training and validation metrics may appear poor, the intermediate results in imgs/ folder indicate that the model is learning. If you do not change any parameters in the model, then, between epoch 400 and 1000, you should start to see that the errors in the intermediate images in folder imgs/ are significantly less. You do not need to train for more than 2000 epochs, assuming you do not change parameters in the current model. |
|
hello Bird9000: Sorry for the late reply. My problem was that I was new to GAN training, and so, I focused on metrics. As training progresses, what you have to do is observe the intermediate results in imgs/ folder. Even though the training and validation metrics may appear poor, the intermediate results in imgs/ folder indicate that the model is learning. If you do not change any parameters in the model, then, between epoch 400 and 1000, you should start to see that the errors in the intermediate images in folder imgs/ are significantly less. You do not need to train for more than 2000 epochs, assuming you do not change parameters in the current model
On Friday, March 25, 2022, 05:33:12 AM CDT, Bird9000 ***@***.***> wrote:
hi ,I got that same problem even I trained for 10000 epochs after 4000 epochs nothing much change . Can i have some tricks
and you say you have pre-trained model that trained for 2000 epochs where will i find it.
—
Reply to this email directly, view it on GitHub, or unsubscribe.
You are receiving this because you authored the thread.Message ID: ***@***.***>
|
|
thank you for the information @tumis1946 @hendraet |
Thanks very much for sharing your work with us.
I wish to ask for your help. I have trained the model from scratch for more than 4000 epochs, but, it is not learning. I did not change anything in your code except the directory for IAM. The pre-trained model that you shared was trained for 2000 epochs and it is a thousand times better than the model I trained for 4000 epochs.
Please, let me know if I have to make any changes in your code for learning to take place.
Thanks
The text was updated successfully, but these errors were encountered: