ValueError: NumpyArrayIterator is set to use the data format convention "channels_last" about dccm HOT 4 CLOSED

Hi, sorry for the late reply. You could change the "image_data_format" from channels_last to channels_first in you Keras configuration file. (usually, it's at $HOME/.keras/keras.json) I think that would help to solve the issue.
Let me know if bugs still exist.

from dccm.

Hi, sorry for the late reply. You could change the "image_data_format" from channels_last to channels_first in you Keras configuration file. (usually, it's at $HOME/.keras/keras.json) I think that would help to solve the issue.
Let me know if bugs still exist.

thanks for your reply, but this leads to another error...
it shows:

Traceback (most recent call last):
File "main_2.py", line 309, in
main()
File "main_2.py", line 140, in main
train(dataloader, model, dim_loss, crit_label, crit_graph, crit_c, optimizer, epoch, datagen, tb_logger)
File "main_2.py", line 188, in train
input_bs = input_tensor[address]
RuntimeError: index 921 is out of bounds for dim with size 855

I think the reason for this error is the setting of 'address', or the 'large_bs' and the 'small_bs'.
specifically,i set:

large_bs: 1000
small_bs: 16
repeated: 10
input_size: 32
workers: 4
num_classes: 4

so i wonder that something wrong in my above setting.
should i change large_bs or num_classes?
looking forward to your reply!

from dccm.

you could simply add "drop_last=True" in torch.utils.data.DataLoader at line 92 in the original main.py : ) also refer to the other closed issue at #2 is necessary.

from dccm.

thanks for your kindly help
there's no any error during running period!

from dccm.