ubuntu 20.04 cuda 11.7 python 3.8.10
please go to https://drive.google.com/drive/folders/1uPKYA39GaigZGAuTakJQuCHt10RbRC6s?usp=sharing to download pretrain_model folder and result_model folder to the root directory of this repo
- create .venv environment
python -m venv .venv
source .venv/bin/activate- use pip to install required package
pip install numpy==1.22.4
pip install wheel
pip install yapf==0.40.1
pip install torch==2.0.1+cu117 torchvision==0.15.2+cu117 torchaudio==2.0.2 --index-url https://download.pytorch.org/whl/cu117
pip install -r requirements.txt
pip install torchmetrics
pip install ipykernel
pip install opencv-python- compiling CUDA operators
cd models/dino/ops
python setup.py build install
# unit test (should see all checking is True)
python test.py
cd ../../..- put train, validation, test provided by hw1 in ./data directory and name them as below
/data
|--/annotations
| |--instances_train2017.json
| |--instances_val2017.json
|--/test
|--/train2017
|--/val2017
- put swin backbone and pretrain checkpoints in ./pretrain_model
/pretrain_model
|--checkpoint0029_4scale_swin.pth
|--swin_large_path4_window12_384_22k.pth
- open inference_and_visualization.ipynb and press run all and then it will generate Output_json_for_test.json if you want to recreate my output result. Remember to use .venv as kernel
- open scripts/DINO_train_swin.sh
coco_path=$1
backbone_dir=$2
export CUDA_VISIBLE_DEVICES=$3 && python main.py \
--output_dir logs/DINO/R50-MS4 -c config/DINO/DINO_4scale_swin.py --coco_path $coco_path \
--options dn_scalar=100 embed_init_tgt=TRUE \
dn_label_coef=1.0 dn_bbox_coef=1.0 use_ema=False \
dn_box_noise_scale=1.0 backbone_dir=$backbone_dir \
--pretrain_model_path pretrain_model/checkpoint0029_4scale_swin.pth \
--finetune_ignore label_enc.weight class_embedreplace output_dir logs with other name like test_logs/DINO/R50-MS4
- run below command in this homework root directory
bash scripts/DINO_train_swin.sh ./data ./pretrain_model 0
- run using evaluate.py
python evaluate.py ./val_result.json ./data/annotations/instances_val2017.json
python evaluate.py output.json [test_truth_path]