search

jazzsaxmafia / show_attend_and_tell

10 stars 6 forks source link

model.npz.pkl for generate_caps.py #3

Open ghost opened 8 years ago

ghost commented 8 years ago

As seen in the main function, the method requires a model, which is a dictionary with all the hyperparameters. I've tried stretching to all the possible parameters, but still i'm missing some of them. Can you please share the one that you have used in your flickr8k / flickr30k experiments? Thanks!

jazzsaxmafia commented 8 years ago

Hi. Actually this not official source code that the author of the paper wrote. Why don't you check the official one https://github.com/kelvinxu/arctic-captions

Or, I wrote a working version in TensorFlow https://github.com/jazzsaxmafia/show_attend_and_tell.tensorflow

Thank you -Taeksoo

ghost commented 8 years ago

Thanks for the swift reply. Actually this is a part of their code, in generate_caps.py 

def main(model, saveto, k=5, normalize=False, zero_pad=False, n_process=5, datasets='dev,test', sampling=False, pkl_name=None):
    # load model model_options
    if pkl_name is None:
        pkl_name = model
    with open('/home/f0z/feat/flickr30k/model.pkl', 'rb') as f:
        options = pkl.load(f)

The model is actually the name of the options pkl file. There are some routines for building / initializing one partially in capgen.py, but I can't get what all parameters are needed. Can you please take a look at them both? Thanks!

BTW, great work on the tensorflow version.