PyTorch implementation of our DHVC 1.0 (AAAI 2024)
- Python 3.8+
- CUDA 11.0
- pytorch 1.11.0
- For others, please refer to requirements.txt
The pretrained models of DHVC 1.0 can be downloaded from NJU Box.
- Train dataset: Vimeo90k
- Test dataset: UVG、MCL-JCV、HEVC Class B
Please download the pretrained models and configure the environment properly first.
Follow the command below to run testing in the dhvc-1.0 folder:
python test.py -d test_dataset_name -c checkpoint_path -p test_dataset_path -g 32 -f 96 -d represents the name of the test dataset used in log file. -c, -p represent the path of the pretrained models and test dataset. -g, -f represent the GOP size and total frame numbers for evaluation. By default, the pretrained models will be placed in ./pretrained, the test dataset will be placed in ./dataset. The test results can be found in ./runs.