search

yanzq95 / SGNet

SGNet: Structure Guided Network via Gradient-Frequency Awareness for Depth Map Super-Resolution (AAAI-2024)
Apache License 2.0
94 stars 3 forks source link

Question about training data #15

Open Ian0926 opened 1 week ago

Ian0926 commented 1 week ago

Thanks for your great work. NYU data contains 2,000 training images. Are only the first 1,000 images used for training in your work?

yanzq95 commented 1 week ago

The NYU dataset contains 1,000 pairs of RGB-D for training and 449 pairs for testing. In this paper, we use the training dataset preprocessed by DKN, where each image (480×640) is evenly divided into TWO (480×320) images along the width dimension. Therefore, there are 2,000 pairs of RGB-D included in "train_xxx.npy" in total.

Ian0926 commented 1 week ago

Thanks for your response.