train/test/val.txt

[FYY799]

Hi, thank you for your great work, your code is very excellent! But may I ask you for some questions about the train.txt/test/val.txt?Can you provide the the files above?

[youngfly11]

You can simply pair each image with their caption sequentially by yourself. like xxxx1.jpg, 0; xxxx1.jpg 2; xxxx1.jpg, 3; xxxx2.jpg, 0, xxxx2.jpg 1; xxxx2.jpg 2.

[FYY799]

I'm sorry for bothering you again.
我下载得到了train/test/val.txt，但是其中只有img_id而没有sent_id,请问sent_id该怎么获取呢？

[youngfly11]

the sent_id is just what we generate, which 0, 1,2,3,4 indeed; This id can index the caption in sent_anno.json

[FYY799]

This is my train.txt:
5517849007 0
5221111799 3
3363750526 2
7014591581 1
2137836768 2
299265726 1
3323952123 0
4108989020
4956585720
4235671794
2158925258
4259870068
4510049538
2405325546
384465575
4791487303

请问每个img_id后的sent_id是随机的0,1，2,3，还是这个index是有要求的？

[youngfly11]

like this:
5517849007 0
5517849007 1
5517849007 2
5517849007 3
5517849007 4
5221111799 0
5221111799 1
5221111799 2
5221111799 3
5221111799 4
....
We generate the train.txt like this because each image corresponds to 5 captions in Flickr30k

[FYY799]

谢谢！

[FYY799]

flickr_datasets/flickr30k_feat_nms/flickr30k_res50_nms1e3_feat_pascal/4398362068.pkl
请问这些pkl文件是如何获得的？