视频质量增强模型加速算法

Video Quality Enhancement Model Acceleration Algorithm

提出了一种应用于视频质量增强算法的动态结构性剪裁算法Maskcut,它可以有效提高基于深度学习的视频质量增强算法的运行速度。Maskcut是一种通用的剪裁思路,支持绝大多数的基于卷积神经网络(CNN)深度学习网络模型的剪裁加速。基于原模型中已经训练好的参数数据,Maskcut使用一种针对剪裁加速的二次训练策略来进一步微调参数,从而在保证模型有效性损失不大的同时,缩短模型运行时间。以一种先进的视频质量增强算法——多帧质量增强2.0(MFQE 2.0)为目标,Maskcut剪裁后可以快速达到峰值信噪比(PSNR)指标损失低于1%、时间缩短10%以上的加速指标。

Maskcut,a dynamic structural clipping algorithm for video quality enhancement is proposed,which can effectively improve the speed of video quality enhancement algorithm based on deep learning.Maskcut is a general tailoring idea that supports most of the tailoring acceleration based on the deep learning network models for convolutional neural networks(CNN).Based on the trained parameter data in the original model,the secondary training for tailoring acceleration is carried out to further fine-tune the parameters.With an advanced video quality enhancement algorithm,the multi-frame quality enhancement 2.0(MFQE 2.0)as the goal,the peak signal-to-noise ratio(PSNR)index is less than 1%and the time is shortened by more than 10% after Maskcut clipping.