A Parameter-Efficient Adaptation of Pre-Trained Models for Action Recognition ...
Wele Gedara Chaminda Bandara, Vishal M Patel
Johns Hopkins University
Accepted at FG'24
Comparison of our Attention Prompt Tuning (APT) for videos action classification with other existing tuning methods: linear probing, adapter tuning, visual prompt tuning (VPT), and full fine-tuning.
Attention Prompt Tuning (APT) injects learnable prompts directly into the MHA unlike VPT.
Setup the virtual conda environment using the environment.yml
:
conda env create -f environment.yml
Then activate the conda environment:
conda activate apt
We use VideoMAE pretrianed on Kinetics-400 dataset for our experiments.
The pre-trained models for ViT-Small and ViT-Base backbones can be downloaded from below links:
Method | Extra Data | Backbone | Epoch | #Frame | Pre-train |
---|---|---|---|---|---|
VideoMAE | no | ViT-S | 1600 | 16x5x3 | checkpoint |
VideoMAE | no | ViT-B | 1600 | 16x5x3 | checkpoint |
If you need other pre-trained models please refer MODEL_ZOO.md.
We conduct experiments on three action recognition datasets: 1) UCF101 2) HMDB51 3) Something-Something-V2.
Please refer DATASETS.md for access to those links and pre-processing steps.
We provide example scripts to run the attention prompt tuning on UCF101, HMDB51, and SSv2 datasets in scripts/
folder.
Inside scripts/
you can find two folders which corresponds to APT finetuning with ViT-Small and ViT-Base architectures.
To fine-tune with APT you just need to execute finetune.sh
file -- which will launch the job with distributed training by
For example, to fine-tune ViT-Base on SSv2 with APT, you may run:
sh scripts/ssv2/vit_base/finetune.sh
The finetune.sh
looks like this:
# APT on SSv2
OUTPUT_DIR='experiments/APT/SSV2/ssv2_videomae_pretrain_base_patch16_224_frame_16x2_tube_mask_ratio_0.9_e2400/adam_mome9e-1_wd1e-5_lr5se-2_pl2_ps0_pe11_drop10'
DATA_PATH='datasets/ss2/list_ssv2/'
MODEL_PATH='experiments/pretrain/ssv2_videomae_pretrain_base_patch16_224_frame_16x2_tube_mask_ratio_0.9_e2400/checkpoint.pth'
NCCL_P2P_DISABLE=1 OMP_NUM_THREADS=1 CUDA_VISIBLE_DEVICES=0,1,3,4,5,6,7,8 python -m torch.distributed.launch --nproc_per_node=8 \
run_class_apt.py \
--model vit_base_patch16_224 \
--transfer_type prompt \
--prompt_start 0 \
--prompt_end 11 \
--prompt_num_tokens 2 \
--prompt_dropout 0.1 \
--data_set SSV2 \
--nb_classes 174 \
--data_path ${DATA_PATH} \
--finetune ${MODEL_PATH} \
--log_dir ${OUTPUT_DIR} \
--output_dir ${OUTPUT_DIR} \
--batch_size 8 \
--batch_size_val 8 \
--num_sample 2 \
--input_size 224 \
--short_side_size 224 \
--save_ckpt_freq 10 \
--num_frames 16 \
--opt adamw \
--lr 0.05 \
--weight_decay 0.00001 \
--epochs 100 \
--warmup_epochs 10 \
--test_num_segment 2 \
--test_num_crop 3 \
--dist_eval \
--pin_mem \
--enable_deepspeed \
--prompt_reparam \
--is_aa \
--aa rand-m4-n2-mstd0.2-inc1
Here,
OUTPUT_DIR
: place where you wants to save the results (i.e., logs and checkpoints)DATA_PATH
: path to where the dataset is storedMODEL_PATH
: path to the downloaded videomae pre-trained model- specifiy thich gpus (gpu ids) you wants to use for finetuning in
CUDA_VISIBLE_DEVICES=
... nproc_per_node
is the number of gpus using for fine-tuningmodel
is the vit-base (vit_base_patch16_224) or vit-small (vit_small_patch16_224)transfer_type
specifies which finetuning method to use. 'random' means random initialization, 'end2end' means full end-to-end fine tuning, 'prompt' means APT (ours), 'linear' means linear probingprompt_start
refers to starting trasnformer block where you add attention prompts. 0 means you start adding learninable prompts from 1st transformer block in vitprompt_end
refers to ending trasformer block where you stop adding attention prompts. vit-base / vit-small has 12 transformer blocks. hence 11 here means you add prompts until last trasnformer blockdata_set
specifies the dataset-
- all the other parameters are hyperparamters related to apt fine-tuning.
If you think this project is helpful, please feel free to leave a star and cite our paper:
@misc{bandara2024attention,
title={Attention Prompt Tuning: Parameter-efficient Adaptation of Pre-trained Models for Spatiotemporal Modeling},
author={Wele Gedara Chaminda Bandara and Vishal M. Patel},
year={2024},
eprint={2403.06978},
archivePrefix={arXiv},
primaryClass={cs.CV}
}
This repocitory is built on top of VideoMAE: https://github.com/MCG-NJU/VideoMAE codebase and we approcite the authors of VideoMAE for making their codebase publically available.