Error Message for Retraining the SSD7 Model after Restarting the Keras Program about ssd_keras HOT 5 CLOSED

Thanks a lot for great materials here :) I changed the model from SSD7 to SSD300 and set image width and height to 300 as in the original code. Then, retraining the SSD300 model was done well even after I restarted the Keras program. Please, have a nice day!

from ssd_keras.

The input image size got messed up at some point, which results in different spatial dimensions in all layers of the network, in particular of the predictions layer. I don't know how exactly you arrived at this point, but 10316 is exactly the number of boxes that the model outputs with the default settings in train_ssd7.ipynb with the preset image size of (300, 480).

I realize that it is not ideal that the predictor_sizes variable is hard-coded in the code cell in section 1. If you change the image size, the predictor sizes will also change, so the way the notebook is currently set up is error-prone. I'll change that.

from ssd_keras.

The potential source of this problem has been fixed in the latest commits and it should now no longer occur, at least not if you follow the instructions in the notebook.

from ssd_keras.

Thanks a lot for your kind help and great materials here ^_^ I will always keep your suggestion above in mind :) Please, have a nice day!

from ssd_keras.

I'll close this for now.

from ssd_keras.