Steps to do Multi Sentence Grounding

[anilbatra2185]

hi @zjr2000, @ttengwang ,

Thank you for sharing the work and code. I am unable to find the steps to perform multi-sentence grounding on own dataset. The config files for MSVG (assuming multi-sentence video grounding, link) is using following files which are missing in the repo.

train_proposal_file: data/generated_proposals/dbg_trainval_top100.json
eval_proposal_file: data/generated_proposals/dbg_trainval_top100.json

Appreciate your response.

[zjr2000]

hi @zjr2000, @ttengwang ,

Thank you for sharing the work and code. I am unable to find the steps to perform multi-sentence grounding on own dataset. The config files for MSVG (assuming multi-sentence video grounding, link) is using following files which are missing in the repo.

train_proposal_file: data/generated_proposals/dbg_trainval_top100.json
eval_proposal_file: data/generated_proposals/dbg_trainval_top100.json

Appreciate your response.

Sorry for the late response and thanks for your attention. For multi-sentence grounding, the main step is to set these two arguments to true:

enable_sentence_context_modeling: True
enable_sentence_pos_embedding: True

This is to enable the interaction between multiple sentence queries. And the ```train_proposal_file''' is not used in our current code (we deleted it in the previous version)