peterzpy / TAL

Faster rcnn architecture for Action Location

Temporal Action Location

There are still many problems to be solved, it's not a mature version!!!

Faster rcnn architecture for Action Location

Install

1. git clone [e-mail]:peterzpy/TAL.git
2. cd TAL
3. mkdir Pretrained
4. cd utils && python3 setyp.py nms/setup.py build_ext --inplace

If you want to use pretrained resnet model Resnet3D

Test

1. cd model
2. python3 test_resnet_learn_nms_Net.py --image_path Your/Image/path --annotation_path Your/Annotation/Path --checkpoint_path Your/Checkpoint/Path

Train With Preprocessing

1. cd model
2. python3 train_resnet_learn_nms_Net.py --use_resnet_pth 'True' --pth_path "../Pretrained/resnet-101-kinetics-ucf101_split1.pth" --preoprocess 'False' --image_path Your/Image/path --annotation_path Your/Annotation/Path --checkpoint_path Your/Checkpoint/Path --video_path Your/Video/path --video_annotation_path Your/VideoLabel/path --feature_preprocess True --feature_path Your/feature/path

Train Without Preprocessing

1. cd model
2. python3 train_resnet_learn_nms_Net.py --image_path Your/Image/path --annotation_path Your/Annotation/Path --checkpoint_path Your/Checkpoint/Path --feature_path Your/feature/path