Open NguyenVietPhong opened 1 year ago
Hello, I have been training under flickr8k data set, the convergence has not been, do you know what the problem is?
Hello, I have encountered with the same problem when I train the flickr8k and flickr30k dataset in my own environment. However, I can correctly run caption.py when I use the best coco_checkpoint the author provided in README. Could anyone help? Thanks!!!
metric blue-4 doesn't increase more than 0.147 with coco data and 0.099 with flickr8k data, is there any way to improve the result. thank