Thank you for your work! I have some questions regarding the process of validatibf the model performance. The experimental results means that you don't validate on the original images, but on the cropped images. Additionally, techniques such as multi-scale and flips approaches seem to have not been applied, which differs from other semantic segmentation methods.
PS: If i only want to validate the model without trainning, should i just add --test and --ckpt xxxx?
Thank you for your work! I have some questions regarding the process of validatibf the model performance. The experimental results means that you don't validate on the original images, but on the cropped images. Additionally, techniques such as multi-scale and flips approaches seem to have not been applied, which differs from other semantic segmentation methods.
PS: If i only want to validate the model without trainning, should i just add --test and --ckpt xxxx?