Inspired by the rapid advancements in deep learning, we propose in this paper a wide-activated squeeze-and-excitation deep convolutional neural network (WSE-DCNN) technique-based video quality enhancement for VVC. Therefore, we replace the conventional in-loop filtering in VVC by the proposed WSE-...
2, for coding each frame xt with frame index t, our coding pipeline contains three core steps: fmotion, fT context, and fframe. At first, fmotion uses optical flow network to es- timate the motion vector (MV) vt, then vt is encoded and decoded as vˆt. Second, based on vˆt...
3. Method We first describe the architecture, block-based warping scheme, and training losses needed to train a 32-bit floating point model. We then describe the quantization procedure, and how we run entropy coding and inference on-device. 3.1. Network ...
JVET Common Test Conditions and Evaluation Procedures for Neural Network Based Video Coding Technology; JVET-AC2016-v1; Joint Video Experts Team (JVET): Geneva, Switzerland, 2023. [Google Scholar] ffmpeg Documentation. Available online: https://ffmpeg.org/ffmpeg.html (accessed on 22 March 2023)...
This enhancement demonstrates the potential of self-attention mechanisms to revolutionize post-filtering techniques in video coding beyond the limitations of convolution-based methods. The experimental results show that the proposed network achieves an average BD-rate reduction of 10.40% for the Luma ...
3. Deep-Learning-Based Image Compression Model (DSSLIC) Figure 2demonstrates the overall architecture of the model used in our work, which is derived from the DSSLIC model already introduced in [2]. As a key distinction to the original model, here we do not use a segmentation network in ou...