Clarioooo
Clarioooo
I found that the web is changed to (https://homes.cs.washington.edu/~ranjay/visualgenome/api.html) instead of (visual genome.org). Which one should I use in VQA? Part 1 or Part2? And which version?
I may have figure out what is wrong. In the bert_finetune.py's def finetune_mode() parameter num_labels should be altered. If it's 12, "Target 12 is out of bounds" as b_labels includes...
Every first three lines is # xxxx. But the code nees the type int just as line4.
Should I write another python file to preprocess the original data or just run the train.py to preprocess it?
I found the generated json file in the /pos_tags/slu/tag_cache/en/ directory. It looks like this