python>=3.5
pytorch>=1.0.0
opencv
numpy
The hardware environment is NVIDIA GTX 1080Ti.
We need to prepare 'file_list.txt', the data format inside is as follows:
/data/oulu/spatial/Train_files/1_1_01_1_avi/frame4.jpg /data/oulu/depth/Train_files/1_1_01_1_avi/frame4.jpg 1
the /data/oulu/spatial/Train_files/1_1_01_1_avi/frame4.jpg is the location of the RGB picture, and /data/oulu/depth/Train_files/1_1_01_1_avi/frame4.jpg is the position corresponding to the Depth picture. 1 is label.
You can modify the configuration according to your needs by modifying options.py . And run :
python main.py
to start a train.
The directory of the trained model is determined by --checkpoints_dir and --name in options.py.
The --model in options.py represents three different models, you can get specific details from the ablation experiment of the paper.
If there is a CUDA out of memory problem, you can modify gpu_ids or batch_size in options.py.
We provide three models in the checkpoints. You can run:
python test.py --name model1
python test.py --name model2
python test.py --name model2
to test different models.