tnti/lmve

LEARNING-BASED MULTI-FRAME VIDEO QUALITY ENHANCEMENT

LEARNING-BASED MULTI-FRAME VIDEO QUALITY ENHANCEMENT

Junchao Tong*, Xilin Wu*, Dandan Ding*, Zheng Zhu**, Zoe Liu**
* Hangzhou Normal University
** Visionular Inc.

---

• Example of the subjective quality for the HEVC compressed video(the 1st column and the 2nd column) and the preprocessed frames(MF、HF).

  Here, MF is the Moderate-quality Frame and HF is the High-quality Frame.

---

• Subjective quality performance on FourPeople and BasketballPass at QP=37 in RA configuration.

---

The experiment section

1.Performance of our proposed LMVE in comparison with LSVE and HEVC in terms of PSNR(dB).

2.Performance of LMVE Compared with MFQE at QP=37.

• The architecture of the QE-subnet is shown in Figure 6, and the details of the convolutional layers are presented in Table 3.

3.PSNR(dB) results of each frame in the first 21 frames on the FourPeople and PeopleOnStreet sequences.

• For more details on the experiment section, you can download the "experiment_tjc.xlsx" file in the current web page

---

The open source address for FlowNet is: https://github.com/lmb-freiburg/flownet2
The open source address for MFQE is: https://github.com/ryangBUAA/MFQE

---

Thanks for reading!
best wishes,
Junchao Tong.