search

Ziyang412 / UCoFiA

Pytorch Code for "Unified Coarse-to-Fine Alignment for Video-Text Retrieval" (ICCV 2023)
https://arxiv.org/abs/2309.10091
MIT License
61 stars 0 forks source link

Issues with unpacking the dataloader when evaluating t2v #10

Closed HuJW0310 closed 2 months ago

HuJW0310 commented 2 months ago

Hi, I'm running the ./eval_t2v/main_ucofia.py on MSRVTT and Something-something-v2 (I modified the dataloader for the latter, and it worked correctly on training) and I got the ERROR listed below:

Traceback (most recent call last):
  File "main_ucofia.py", line 664, in <module>
    main()
  File "main_ucofia.py", line 661, in main
    eval_epoch(args, model, test_dataloader, train_dataloader, device, n_gpu)
  File "main_ucofia.py", line 450, in eval_epoch
    input_ids_train, input_mask_train, segment_ids_train= batch
ValueError: too many values to unpack (expected 3)

(I modified the code a little bit so the lines doesn't match, just ignore it pls)

I noticed that there's a comment says "Maybe something went wrong here!!!" right at the malfunctioning function eval_epoch() (in file "./eval_t2v/main_ucofia.py", line 404 & file "./train/main_ucofia.py", line 363, from the original code in the repo)

I wonder if you have figured out what's wrong with it & the corresponding solution