Error while running eval.py on pretrained models

[KahHwa]

Hello,

This seems like a stupid question but why am i getting this error when i run the eval.py on the top down model. Does the code not work on python3?

UnicodeDecodeError: 'ascii' codec can't decode byte 0x84 in position 0: ordinal not in range(128)

[KahHwa]

I used pickle.load(f, encodings='latin1') instead of the cPickle.load(f). WIll this work instead ? All the deprecated functions, I have changed them as well so that I can generate captions from the pretrained models.

So far the results are reasonable for resnet101 but with resnet152 the captions are just totally off. Why is the output so different?

[08tjlys]

hi, @KahHwa i think you r right, i also found caption generated with resnet152 were totally not reasonable, but with resnet101 they were ok. it seems there are something wrong with the pretrained resnet152 author provided

[ruotianluo]

Because the models are trained with resnet101 feature. If you want to use rednet152, you need to train on your own.

[08tjlys]

thanks for that information!