miracle24

Hello, @zhunzhong07. I encountered a similar problem. I also wonder why to use q_q_dist? I mean, in practical ReID scenarios, there may be not enough query images for q_q_dist.

Hello. I found in caption_all.json, no split information is provided for each image. @bunKiatIunn @ShuangLI59 Do you know how to split the dataset into the training set, validation set, and...

Thanks! Another thing. I used my own preprocessing code for the box information (eg. getting region index, computing the percentage of the overlapping area, normalizing the overlapping weights) and your...

Actually, when I use pytorch0.4, label 'S-Stop' (the corresponding id in SCRF_l_map is 5) may be in the middle of the decoded_scrf, instead of in the end of it. See...

Did you fine-tune the CNN when trained the model with cross entropy loss?

Wow. It's unbelievable. I can not achieve that high score without fine-tune when train my own captioning model under cross entropy loss. Most papers I have read will fine-tune the...

I think they means they did not do finetuning when trained the model under RL loss, while they did not mention whether they finetune the CNN when train the model...

I finetnue the CNN under cross entropy loss as neuraltalk2 (Lua version) and I got cider 0.91 on validation set without beamsearch. Then I train the self-critical model without finetuning...

I did not train the attention based model. But I will try. Thank you and your codes. I will start learning pytorch with you code.