Skip to content

Latest commit

 

History

History
82 lines (62 loc) · 4.15 KB

README.md

File metadata and controls

82 lines (62 loc) · 4.15 KB
Raw

OSTrack

Install the environment

conda create -n ostrack python=3.8
conda activate ostrack
bash install.sh

Set project paths

Run the following command to set paths for this project

python tracking/create_default_local_file.py --workspace_dir . --data_dir ./data --save_dir ./output

After running this command, you can also modify paths by editing these two files

lib/train/admin/local.py  # paths about training
lib/test/evaluation/local.py  # paths about testing

Pre-Training

Download pre-trained MAE ViT-Base weights (different pretrained models can also be used, see MAE for more details). Path to the pretrained weights is to be specified in config file MODEL.PRETRAIN_FILE.

python tracking/train.py --name  experiment_name --script ostrack --config vitb_256_mae_ce_32x4_got10k_ep100 --save_dir /path/to/save/directory  --mode multiple --nproc_per_node 1 --use_wandb 1

Replace --config with the desired model config under experiments/ostrack. We use wandb to record detailed training logs, in case you don't want to use wandb, set --use_wandb 0. To use wandb, add auth key in config TRAIN.wandb_key which can be found here.

Pretrained OSTrack weights can be downloaded from here.

Sparsity Training

Specify path to OSTrack pretrained weights in config file MODEL.PRETRAIN_FILE. Make sure to set sparsity_train in config file as True for sparsity training. W1 and W2 in config are tunable sparsity hyperparameters.

python tracking/train.py --name  experiment_name --script ostrack --config vitb_256_mae_ce_32x4_got10k_ep100_sparse --save_dir /path/to/save/directory  --mode multiple --nproc_per_node 1 --use_wandb 1

Pretrained Sparsity trained weughts can be downloaded from here.

Child Finetune

Add path to sparsity trained weights in config file MODEL.PRETRAIN_FILE. Specify Pruning type (naive or layerwise) and MLP and Attention budgets in config. (Layerwise pruning is generally done in cases of extreme pruning i.e. 1% budget in our case.)

python tracking/train.py --name  experiment_name --script ostrack --child_train 1 --config vitb_256_mae_ce_32x4_got10k_ep100_sparse --save_dir /path/to/save/directory  --mode multiple --nproc_per_node 1 --use_wandb 1

Evaluation

Download the Child finetuned weights from Google Drive

Specify the weight path in config TEST.PRETRAIN_FILE.

Change the corresponding values of lib/test/evaluation/local.py to the actual benchmark saving paths

Some testing examples:

  • OTB
python tracking/test.py  --name result_folder_name   ostrack --tracker_param vitb_256_mae_ce_32x4_got10k_ep100_sparse --dataset otb --threads 6 --num_gpus 1
python tracking/analysis_results.py --results_dir path/to/result/folder --dataset_name otb --config vitb_256_mae_ce_32x4_ep300_sparse
  • LaSOT
python tracking/test.py  --name result_folder_name   ostrack --tracker_param vitb_256_mae_ce_32x4_got10k_ep100_sparse --dataset lasot --threads 6 --num_gpus 1
python tracking/analysis_results.py --results_dir path/to/result/folder --dataset_name lasot --config vitb_256_mae_ce_32x4_ep300_sparse
  • GOT10K-test
python tracking/test.py  --name result_folder_name   ostrack --tracker_param vitb_256_mae_ce_32x4_got10k_ep100_sparse --dataset got10k_test --threads 6 --num_gpus 1
python lib/test/utils/transform_got10k.py --name result_folder_name --tracker_name ostrack --cfg_name vitb_256_mae_ce_32x4_got10k_ep100_sparse
  • TrackingNet
python tracking/test.py  --name result_folder_name   ostrack --tracker_param vitb_256_mae_ce_32x4_got10k_ep100_sparse --dataset trackingnet --threads 6 --num_gpus 1
python lib/test/utils/transform_trackingnet.py --name result_folder_name --tracker_name ostrack --cfg_name vitb_256_mae_ce_32x4_got10k_ep100_sparse