lyricpoem

use loss.item() instead of loss.data[0]

I solve this problem by deleting all images whose size > 400*160(about 250 images) from training set.

运行eval.py的时候加个--spice就行了

是的

Thanks for your interest. As you can see, there are five kinds of keys in our .hdf5 file. They are * ['%d_features' % image_id]: region features (N_regions, feature_dim) * ['%d_boxes'...

yes, it is right

您好，这个问题主要是pytorch版本导致的，pytorch1.10左右的版本开始会把代码里的dataset识别为IterableDataset的子类，所以建议用之前的版本，最好是用1.2，当然1.7之前应该也都可以

I have not got this error. What is the output of print(scores) ?

I download m2transformer pretrained model and test it on karpathy test set. Nearly 1/3 (1718/5000) sentences have bad enddings like ``` 'a woman eating a cupcake with lit candles on...

Yes, I have tried to add eos and trained a model. There are no such sentences. But Cider also decreases. So it seems the only difference between two frameworks?