Does number of torchvision.transforms used affect data loading speed?

As the title says, I’m trying to do some speed test on my data loading code. I am wondering if adding all kinds of transform to transform.Compose() will affect the speed of data loading?

Thank you.

All compose does is apply the transform one by one

for t in self.transforms:
    img = t(img)
return img

So adding all the transforms to compose is same as applying them one-by-one.

1 Like

So if i want to make it as slow, i should add as much transformation as possible? Or creating own augmentation function would be better?

This is true.

Torchvision transforms are written in C++ internally, so they are already fast.

1 Like