Shoufa Chen comments

Results 121 comments of


                                            Shoufa Chen

Pre-training code of BLIP-2

Hi, @dxli94 , Thanks for your awesome work. I was wondering why there is only `pretrain_stage2.sh`? Can I just replace `pretrain_stage2.yaml` with `pretrain_stage1.yaml` for stage 1 pretraining?

How to solve the problem of loss NAN?

Hi, Which configuration do you use? Full-tuning baseline or AdaptFormer?

How to solve the problem of loss NAN?

I am sorry I didn't experiment with your specific configuration. Reduce the learning rate and have a try.

Hi, Thanks for your interest. Good catch. There are some checkpoints from VideoMAE that include `module` instead of `model`. We provide our converted weights. You can find them [here](https://github.com/ShoufaChen/AdaptFormer/releases/tag/v0.1).

Vit-B IN21K weights

Hi, Please find the checkpoint at https://github.com/ShoufaChen/AdaptFormer/releases/download/v0.1/vit_base_patch16_224_in21k_to_video_tz.pth

Vit-B IN21K weights

Could you give post complete log?

Two questions about the experimental results in Tabel 1 of the paper.

Hi, Thanks for your raised question. 1. May I know your detailed configuration including command and pre-trained weights? 2. Good catch. We are sorry for that typo. We updated the...

Two questions about the experimental results in Tabel 1 of the paper.

Thanks for your reply. Did you experiment with the VideoMAE codebase? I guess you experiment with strong data augmentation and optimizer (e.g. AdamW). For fair comparison to linear probling, We...

[ Preprocessing SSv2 ]

Hi, Please follow https://github.com/MCG-NJU/VideoMAE/issues/62.

Some question about equation (2) in your paper

Hi, Thank you for your feedback. I apologize for the delayed response due to the holiday season. In our work, we employ the posterior distribution q(z_{t-1} | z_t, z_0) to...