Open ShinDongHwan opened 6 years ago
Hi Shin Dong, Sorry for late reply, and I'm afraid that I couldn't find the original code to generate the two pickle files now, and I'm currently on vacation. But I think you are correct, they are from that repo, so the two files are key value pairs (image id -- caption). The original repo author has ceased any support, and I'm not working on this model any more, so I would recommend you to move onto other image captioning models. Please check Luo Ruotian's implementation of self-critical model. That repo has more models with better performance. If you still want to evaluate, I will try to find them after vacation. Thanks.
Thank you for your kindly answer. I'm sad to hear that you're not working on it any more. because i want to generate paragraphs, and I couldn't find generating paragraphs model. if you have the rest time, please try to find them after vacation.
Have a good vacation~!!^^
Hi Shin Dong, From my experience, I don't think this model is the only choice for generating paragraphs, and in fact, it also suffers from long term dependency problem. So in other words, it may not outperform other models such as self-critical, or worse, on my private dataset, it achieves lower scores. But I will try to find scripts and push them to GitHub some time later. Thanks.
Oh, Thank you. I was impressed with your advice and help. I began to study about language model, so I lacked knowledge about it. Thank you very much!!
Hi, Wentong.
I am also facing the same problem during the evaluation. As there are not "references.pkl" & "candidates.pkl" at "data" directory. How can I get this? can you please help me out?
Same issue here. Are these files somewhere present? Or maybe at least someone can explain in detail what should be in them?
Hi, Wentong. I finished training. so I want to evaluate a model. but there are not "references.pkl" & "candidates.pkl" at "data" directory.
should i download and execute this link("show-attend-and-tell")? how do i get them?