Video Deblurring¶
cross-reference: Image Deblurring
IFI-RNN¶
Recurrent Neural Networks with Intra-Frame Iterations for Video Deblurring (CVPR 2019)
No source code?
STFAN¶
Spatio-Temporal Filter Adaptive Network for Video Deblurring (ICCV 2019) - SenseTime + Nanjing University of Science and Technology
Project | PyTorch 1.0
- We propose a filter adaptive convolutional (FAC) layer that applies the generated element-wise filters to feature transformation, which is utilized for two spatially variant tasks, i.e. alignment and deblurring in the feature domain.
- We propose a novel spatio-temporal filter adaptive network (STFAN) for video deblurring. It integrates the frame alignment and deblurring into a unified framework without explicit motion estimation and formulates them as two spatially variant convolution process based on the FAC layers.
CDVD-TSP¶
Cascaded Deep Video Deblurring Using Temporal Sharpness Prior (CVPR 2020) - Nanjing University of Science and Technology
Project |
PyTorch code | Author’s blog overview (chinese)
- We propose a simple and compact deep CNN model that simultaneously estimates the optical flow(PWC-Net) and latent frames for video deblurring.
- To better explore the properties of consecutive frames, we develop a temporal sharpness prior to constrain deep CNN models.
Other Note: Motion blur¶
TSP is designed for hand-held cameras dataset from DVD based on below assumption
As demonstrated in [4], the blur in the video is irregular, and thus there exist some pixels that are not blurred. Following the conventional method [4], we explore these sharpness pixels to help video deblurring.
Hence it might not suitable for continuous motion blur (e.g. racing)
BIN¶
Frame interpolation/BIN
EDVR is better than BIN when only deblur