second epoch weights issues

[sujayr91]

Hi,

Looks like the weights starting from the second epoch, gives an issue while testing with sample image.
The output is only sequence of start ...
When i debugged, I see that the "outputs" tensor array in model.py contains all "nan"

[yunjey]

@sujayr91 I don't understand what you are asking.. What do you mean by 'weights starting from the second epoch'?

[sujayr91]

Hi yunjey,

The model parameters I am talking are with respect to files which will be saved in model/ during training epochs. First I trained the model for Coco dataset. While testing sample.py, if i choose model parameters any where starting from second epoch (for example encoder-2-1000.pkl , and decoder-2-1000.pkl) I get output caption as .......

I will check your pretrained model also.

Thanks

[yunjey]

Oh, I think you should change the code in configuration.py.

I deleted the configuration.py file and add argparse to control the arguments. Please check the code.