Self-Chained Image-Language Model for Video Localization and Question Answering
Please follow the official code to reproduce the results on NExT-GQA. Afterwards, you can use the tools provided here to convert the predicted frame ids into timestamps for evaluation. To run the code, you need to copy tools into sevila and change the file paths accordingly.
python fid2seconds.py
python split_qa_los.py