This code has been released for the RGBD Triathlon benchmark.
It's a script to adapt the database format to be loaded by PyTorch ImageFolder Dataset class.
It's the script to run the toy example. You can start from here to add your method or evaluate methods.
It's the code to perform the training. The training hyperparameters are:
- Random resizing crop every image to 224*224 and normalization (with imagenet mean and std)
- Optimizer is SGD (default are lr=10e-3 and momentum=0.9)
- Scheduler is set to decrease by 10 the lr after 40 steps
- loss function is Cross Entropy