Open joez17 opened 1 year ago
And when i use the given 5M checkpoint to finetune on msrvtt-9k retrieval task, the result is lower than given results. Fro example, R@1 is 39.7 for best.
Did you run the evaluation code after fine-tuning? The validation results during finetuning are evaluated with 4 frames for faster training. We evaluate our final results using 12 frames.
Did you run the evaluation code after fine-tuning? The validation results during finetuning are evaluated with 4 frames for faster training. We evaluate our final results using 12 frames.
That's right! Thanks! And can you provide pretraining logs for further reference?
Hi! I'm trying to pretrain VindLU using 5M data, can you provide the pretraining logs for reference? Thanks!