EDVR Video-Enhancement and Super-Resolution in Keras

EDVR: Video Restoration with Enhanced Deformable Convolutional Networks

Keras implementation of "Video Restoration with Enhanced Deformable Convolutional Networks" CVPR, 2019: EDVR Based on author's Pytorch implementation

Model Architecture

TODO:

Help needed in translation of authors' DFA module to Keras\Tensorflow: DCN_V2

• [x] PreDeblur Module
• [ ] Alignment with Pyramid, Cascading and Deformable Convolution (PCD) Module
• [x] Temporal Spatial Attention (TSA) Fusion Module
• [x] Reconstruction Module
• [x] Pixel shuffle layer (Subpixel)
• [ ] Support changing frame shape (currently supports only fixed shape)

Results

• Dependencies

  • Python 3.6
  • Keras>2.2.x