Figure 1 | Scientific Reports

Figure 1

From: Stereoscopic video deblurring transformer

Figure 1

The proposed stereo video deblurring model. Firstly, PWC-Net estimates the motion between the neighboring frames. Then, we apply a 3D CNN layer to the motion-compensated frames, and the proposed Transformer model accepts the resulting features as input. Next, another CNN layer (CRB) extracts deep features. The mPAM then fuses the stereo input features. A convolutional decoder constructs the deblurred frames from the left and right features. Finally, we form the output by adding the blurry middle target frames with the reconstructed left and right frames.

Back to article page