Why did you augment the test set as well as the training set? And why did that boost the performance of your proposed method? - Githubissues

rfuruta / pixelRL

135 stars 31 forks source link

Why did you augment the test set as well as the training set? And why did that boost the performance of your proposed method? #14

Open rgeng5 opened 2 years ago

rfuruta commented 2 years ago

Are you talking about this implementation? If so, you might be misunderstanding because test-time augmentation is not implemented in this code.

If you are talking about our paper, please see Sec. 7.2 in this report (https://openaccess.thecvf.com/content_cvpr_2017_workshops/w12/papers/Agustsson_NTIRE_2017_Challenge_CVPR_2017_paper.pdf). We just followed this augmentation technique. The performance was improved by ensembling the enhanced predictions.