search

chuckcho / video-caffe

Video-friendly caffe -- comes with the most recent version of Caffe (as of Jan 2019), a video reader, 3D(ND) pooling layer, and an example training script for C3D network and UCF-101 data
Other
175 stars 93 forks source link

How to use the pre-trained model which was trained on sport 1M dataset? #66

Closed bitwangdan closed 7 years ago

bitwangdan commented 7 years ago

I run the code successfully. the model is pre-trained on the ucf_101 dataset ,but when i replace it with a model pre-trained on the sport1M dataset ,the code was wrong , can you tell me how to deal with it ?

pelun commented 7 years ago

After I train this model, I run plot_training_loss.sh. It shows:

No images found in file: /tmp/iterloss${UUID}.png

Help me resolved this problem, please!

pelun commented 7 years ago

@chuckcho Can you help me?

xuzhenqi commented 7 years ago

I also have the need to use the pre-trained model from sport1M dataset, since it performs better. It's convenient if this repo contains such examples.

pelun commented 7 years ago

@xuzhenqi Can you help me resolved errors above?

chuckcho commented 7 years ago

@pelun your question (https://github.com/chuckcho/video-caffe/issues/66#issuecomment-264759702) is not related to the original question. plot_training_loss is quite hacky, and assumes many things (python packages, directory / filenaming, etc). Look at the file to resolve your case please.

chuckcho commented 7 years ago

@bitwangdan sorry for the late reply. When you say model pre-trained on the sport1M dataset, what exactly is it? Isn't that model trained using facebook/C3D? Then, it's not compatible.

bitwangdan commented 7 years ago

@chuckcho Hi, in this example ,you only offer the code about the ucf101, but the sport 1M hava a good result on the feature extraction , can you offer the code about the sport 1M , i need your help.

chuckcho commented 7 years ago

I'll try to convert sports1M pre-trained model into the format that's useable here. Please bear with me.

LongLong-Jing commented 7 years ago

@chuckcho Can you give me some hint about how to convert Sport1M pre-trained model to the format that I can use here? If I know how to do it, I can try to convert the model. Thanks!

chuckcho commented 7 years ago

i'll do the conversion in my spare time. Moved this issue to TODO list in README.md for now. Thanks.