tsenghungchen / SA-tensorflow

Soft attention mechanism for video caption generation

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

What does the 'feature_path' mean?

lcmaster-hx opened this issue · comments

Hi, I have trouble with what the feature_path means?
As you said, "The path that stores features such as VGG and C3D". Does it means the weights from VGG-16 which is already trained?
I have a pretrained vgg16_weights.h5 file but it doesn't work well.

hmm.. are there no further comments on this? I would also be interested in the answer

I was able to write h5 data generation scripts. I have uploaded my data generation scripts at-
https://github.com/sxs4337/SA-tensorflow/tree/master/prepare_data

See if these help.

WOW thank you very much for the script, I will check it!
Thanks in advance! 👍

commented

@sxs4337 Thank you very much for all this.
Would you be able to share your feature file resnet_feat_all_pool5.pkl or is this too big ?

If this is, could you link us the script you used to extract images and features from the videos ??
tyvm in advance

I remember running the CNN feature extraction in caffe back then. I have uploaded some files that should help.
https://github.com/sxs4337/SA-tensorflow/tree/master/prepare_data

commented

Thank you so much :-]

commented

Does any one has been able to run the code fully ?
I am confused in VGG and C3D features and feature_path.
input_generator has inp_path and feature_path. Can anyone clarify what paths are these ?
thanks