search

YapengTian / TDAN-VSR-CVPR-2020

TDAN: Temporally-Deformable Alignment Network for Video Super-Resolution, CVPR 2020
MIT License
400 stars 62 forks source link

Some questions about the code #4

Closed IanYeung closed 4 years ago

IanYeung commented 5 years ago

https://github.com/YapengTian/TDAN_VSR/blob/c00b8d15f91fbf652102c9fd21c247cdfdbe6e21/model.py#L76-L78

Isn't it the feature aligned in line 76? Why you need to further pass through another deformable convolution layer in line 78?

YapengTian commented 5 years ago

Yes, the features from the deconv_3 in line 76 are aligned features. To further enhance transformation capacity and explore more contexture information for reconstructing the target aligned pixels (the predicted offsets will flexibly enlarge receptive field), we use additional regular deformable convolution layers (not for temporal alignment as the layer: deconv_3).

IanYeung commented 5 years ago

Thanks for your quick response.