soloist97/region-hierarchical-pytorch

Implementation of a baseline method for image paragraph captioning

region-hierarchical-pytorch's Issues

File densecap_image_features_f_max_50_f_size_4096.h5

Hi,

could you please tell us how to get this file that you are using in the train.py program is it the im2p_train_output.h5 file ?

Please let me know.
Thank you.

Hi! Great work! I'm also unable to access it from the above link. Could you please share the files to me, it would be very helpful for me! This is my email: [email protected] . Thanks very much!

How to map bottom-up features in preprocess.py?

Thanks for your excellent code! But I just want to ask how can I map bottom-up features in preprocess.py? I want to know this because I got a Runtime Error when I used the Bottom-Up feature provided by you, which is showned below.(By the way, thanks a lot for providing pretrained features of Dense and Bootom-Up!!) I think this is because the preprocess section throws away some dirty data, but I just skip step 6 in preprocess.py, which causes mismatch. I don't know if this guess is correct.

RuntimeError: Caught RuntimeError in DataLoader worker process 0.
Original Traceback (most recent call last):
RuntimeError: stack expects each tensor to be equal size, but got [52, 2048] at entry 0 and [25, 2048] at entry 

Above all, thanks a lot if you can tell me how to fix this problem!
Thanks for sharing the code and pretrained features again!

the preprocessed files are unavailable

I enter the Onedrive link and got this: "This item might not exist or is no longer available". Could you upload a new one?

Please add results of your model on test set

Hi,

I appreciate your work on creating pytorch version for this project. It would be very helpful if you could also post the results file that you got after training the model. It would easy for us to compare with other models if this file is available. Thank you so much !