This is a modified version of implementing video summarization on custom data. For basic establishment (eg. pretrain-model, TVSUMM & SUMME datasets, etc), please refer to the DSNet source: https://github.com/li-plus/DSNet
The file structure demonstrated below is an example of my arrangement.
I put custom videos in mydatasets/video/
, and the generated h5 file in mydatasets/
.
shot-based-video-synopsis
├── mydatasets/
| ├── video/
| | ├──video1.mp4
| | └──video2.mp4
| └── vsumm_custom.h5
├── output/
└── models/
python data_preparation.py --raw-data /your/video/path --write-h5 /your/h5/path --subsample-size 10
python generate_summary_video.py anchor-based --input-data /your/h5/path --input-video /your/video/path \
--output-folder /your/output/path --model-path /your/model/path