search

Becomebright / GroundVQA

Official PyTorch code of "Grounded Question-Answering in Long Egocentric Videos", accepted by CVPR 2024.
https://arxiv.org/abs/2312.06505
MIT License
49 stars 1 forks source link

Training on Ego4D NLQ. #6

Closed leon-herbrik closed 5 days ago

leon-herbrik commented 6 days ago

Hello,

first of all, amazing work!

I have two questions about the method:

  1. Is it supported to train the model on only the Ego4D NLQ data?
  2. Does the 'annotations.NLQ_train.json' file contain the additional questions you generated with Llama2?

Thank you!

Becomebright commented 6 days ago
  1. Yes, I've uploaded the training and evaluation scripts. You can also find the checkpoints at https://github.com/Becomebright/GroundVQA/issues/2.
  2. No, annotations.NLQ_train.json only contains the training set from NLQv2. The generated samples are in annotations.EgoTimeQA.json.

Let me know if you need any further clarification!

leon-herbrik commented 5 days ago

Got it, thanks a lot :)