Should DVSCifar10 use transform?

[QgZhan]

https://github.com/Gus-Lab/temporal_efficient_training/blob/f9bde4107ed653cc8dd3ee58689bf3b55f6b89ba/data_loaders.py#L107

The code in line 107 is:
train_dataset = DVSCifar10(root=train_path, transform=False)

In line 88 of this file, DVSCifar10 determines whether to use transform using the following code:
if self.transform is not None:

When initializing DVSCifar10, the self.transform is initialized as a bool variable, and self.trainsform is not None will return a False result. I wonder if DVSCifar10 should use transform.

[Jackn0]

Sure, we use the the transform on our paper. Thanks for reminding me of this. I will fix this bug.

[yhhhli]

Hi, the transformation is a kind of data augmentation designed for event-based data. We recently describe such technology in this paper, arXiv link. You could find more details in that paper.