Alapfogalmak
Proposing HFUR for compressed video quality enhancement through frequency-based upsampling and iterative refinement.
Kivonat
The article introduces HFUR, a neural network architecture for enhancing compressed video quality by focusing on frequency-based upsampling and iterative refinement. Video compression artifacts are addressed through the proposed modules: ImpFreqUp and HIR. ImpFreqUp utilizes DCT-domain prior to reconstruct loss, while HIR refines feature maps hierarchically. Extensive experiments demonstrate the effectiveness of HFUR in achieving state-of-the-art performance.
Statisztikák
Extensive experiments show that HFUR achieves state-of-the-art performance.
The proposed method is formulated within a multi-scale framework to mitigate distortions of CUs with varying scales.
In CBR mode, videos are encoded at fixed bit rates of 200kbps and 800kbps.
Compression is conducted with QPs of 27 and 37 in CQP mode.
Idézetek
"Video compression artifacts arise due to the quantization operation in the frequency domain."
"We propose a hierarchical frequency-based upsampling and refining neural network (HFUR) for compressed video quality enhancement."
"HFUR achieves state-of-the-art performance for both constant bit rate and constant QP modes."