Our latest work of video enhancement

Ren Yang, Xiaoyan Sun, Mai Xu and Wenjun Zeng, "Quality-Gated Convolutional LSTM for Enhancing Compressed Video", in IEEE International Conference on Multimedia and Expo (ICME), 2019.

is avaiable at https://github.com/ryangchn/QG-ConvLSTM.git (codes included).

The demo includes the trained model (for HEVC sequences at QP = 37) and test codes of the MF-CNN in our MFQE approach.

In this demo, we use frame 96 (non-PQF) of the video sequence BasketballPass as an example.

As a result, the non-PQF (frame 96) can be enhanced by our MFQE approach, taking advantage of the adjacent PQFs (frames 93 and 97).

Run "./Demo/main.py" to run the demo.

After runing, the frame 96 compressed by HEVC and enhanced by our MFQE approach are shown as "./Demo/Frame_96_HEVC.bmp" and "./Demo/Frame_96_our_MFQE.bmp", respectively.

NOTICE: The trained model (./Demo/HEVC_QP37_model/model.ckpt) are also suitable for non-PQFs of other video sequences compressed by HEVC at QP = 37.

Ubuntu 14.04, Tensorflow 1.3.0, Python 2.7

Tensorflow, TFLearn, Numpy, Scipy, matplotlib, skimage

E-mail: [email protected], [email protected]

WeChat: yangren93