Data augmentation


I read in the documentation that if data augmentation is set up in the spec file, then the image is converted between 0 and 1.

But in my spec file I have set image_scaling_factor: 1.0

In one hand, he scale factor is 1, so image must be in [0, 255].
On the other hand, I activate the dta augmentation, so image must be in [0, 1].

Therefore during inference, should I give an image in [0, 255] or in [0, 1] with this setup ?

Since our input images are scaled between 0 and 1.0, you can set this value to 0.5

Thank you for your help

For Faster_rcnn, the images before mean subtraction and scaling is in the range 0-255 rather than (0, 1) as in DetectNet_v2.