shikiw / dam-vp Goto Github PK

Hi,

Thanks for the amazing work! Would it be possible to share the link for downloading/acquiring your Fru92 and Veg200 data splits? I see that the original dataset is here: https://github.com/ustc-vim/vegfru, but I would like to know the exact processing you've performed on the dataset to get to the two subsets you've used.

Thanks!

Hi!
Could you please provide the full results on VTAB-1k benchmark?
The Table 11 only shows the average results on three types of datasets in VTAB-1k benchmark.
Thanks!

Hi,

Thanks for your excellent work and contributions of open-sourced code.

I think there should be a bunch of json files claiming the train/test splits of downstream datasets. I met this issue when I run

    cd task_adapting
    CUDA_VISIBLE_DEVICES=1 python main.py \
        --base_dir ../data \
        --pretrained_model vit-b-1k \
        --adapt_method prompt_wo_head \
        --test_dataset 'cub200' \
        --epochs 50 \
        --lr 50000 \
        --weight_decay 0.0 \
        --checkpoint_dir ../meta-training/checkpoints/vit-b-1k-wo-head.pth

The error seems lie in a class named 'JSONDataset', which needs 'train.json' to continue training.

Thanks for your efforts.

I want to use this for Video Swin transformer on the Shanghai Video anomaly Dataset , and I want to custom train it for that dataset and model . How can I do it

Hi, This is a good job, but I want to know what kind of GPUs are needed for training.

Hi, thanks for the great work. I realise that you are using AgglomerativeClustering from sci-kit learn to first learn the cluster prototypes. However, in my case, this algorithm keeps giving me a single cluster, do you have any suggestions on this behavior?