This product is an open source software. If you would like to access the source code, please visit the below link. (You do not need to click "ADD TO CART" for this item)

<<To ACCESS Source Code: CLICK HERE>>

Authors: Prof. Qifeng Chen and Chenyang Qi, Junming Chen, Xin Yang

We propose a SOTA streaming video denoising method BSVD that outperforms existing methods on videos with synthetic and real noise in both inference speed and image fidelity. Our pipeline-style inference with Bidirectional Buffer Blocks allows bidirectional temporal fusion for online streaming video processing, which is proved to be more effective than unidirectional fusion. In addition, we solve the degradation of clip edges, which exists in MIMO frameworks. Our method is effective for both non-blind and blind denoising, and is also general for similar architectures. Extensive experiments on public datasets have demonstrated the effectiveness of our method.

Authors:

Department of Computer Science and Engineering Assistance Professor Prof. Qifeng Chen

Key Features:

Our buffer-based pipeline-style inference can be applied to the existing method. FastDVDnet is a two-stage sliding-window-based method that conducts temporal fusion at the input layer of each U-Net. We utilize the pre-trained checkpoint and buffer the intermediate feature during each forward inference, which modifies the original computation graph into pipeline style. As a result, we half the runtime from 42ms to 23ms with the same image fidelity as the original implementation.

Specification:

Please refer to the published paper for specificaion

Reference:

https://cqf.io/papers/Efficient_Video_Denoising_ACMMM2022.pdf

Contact Us:

ttsamuel@ust.hk

Download (1)

Real-time Streaming Video Denoising with Bidirectional Buffers
Efficient_Video_Denoising_ACMMM2022.pdf*
size: 3046396 KB, type: application/pdf
 

You would be able to download the files marked with an asterisk '*' only after logging in to your account and completing the purchase of the product license.