training parameter problems about c3d HOT 5 CLOSED

@TinkerBell123 what do you mean by "the 8 conv networks"? and which dataset you are training on?

from c3d.

I use ucf101 dataset.
when I first train your network,the proto file in c3d_train_ucf101/conv3d_ucf101_train.prototxt has 5 convolutional network layers,I later found the proto files have 8 convolutional layers in finetuning and feature extraction,so I change the 8 conv layers prototxt file to train ucf101.The first I used the small networks,it converge well.But when I change the larger one ,it ocurrs the problem

from c3d.

I know what's the problem,thank you very much

from c3d.

Glad to hear.

from c3d.

hello @TinkerBell123 I have met the same problem like you. I met the problem when I run c3d_train_ucf101/train_ucf101.sh #85. I want to know how did you solve the peoblem. Thank you very much!

from c3d.