torch = 1.7.1+cu110 python = 3.8.13
Some basic python packages such as ninja yacs cython matplotlib tqdm OpenCV scipy ......
Backbone: Resnet-101
| No. of GPUs | mAP | FPS |
|---|---|---|
| 1 | 82.45% | 3.51 |
| 4 | 82.45% | 9.21 |
Backbone: Resnet-50
| No. of GPUs | mAP | FPS |
|---|---|---|
| 1 | 81.28% | 4.46 |
| 4 | 81.28% | 14.10 |
Train the model
python -m torch.distributed.launch --nproc_per_node=1 tools/train_net.py --master_port=$((RANDOM + 10000)) --config-file configs/AFGN/vid_Res_101_C4_AFGN_1x.yaml OUTPUT_DIR training_dir/AFGN
Test the model
Then run
python -m torch.distributed.launch --nproc_per_node=1 tools/test_net.py --config-file configs/AFGN/vid_Res_101_C4_AFGN_1x.yaml MODEL.WEIGHT AFGN_R_101.pth
- The AFGN is based maskrcnn-benchmark and MEGA for medical video object detection.