EDCNN

Efficient In-loop Filtering Based on Enhanced Deep Convolutional Neural Networks for HEVC

The in-loop filtering in HEVC

We propose an efficient in-loop filtering algorithm based on the enhanced deep convolutional neural networks (EDCNN) for significantly improving the performance of in-loop filtering in HEVC. the EDCNN is proposed for efficiently eliminating the artifacts, which adopts three solutions, including a weighted normalization method, a feature information fusion block, and a precise loss function.

---

Our proposed EDCNN

1. The structure of proposed feature information fusion block

<img src="network/20200520103208.png" alt="20200520103208.png" style="zoom: 20%;" />

2. The architecture of proposed EDCNN

3. The detailed network parameters

<img src="network/20200520131415.png" alt="20200520131415" style="zoom:33%;" />

---

Experimental Results

1. The PSNR standard deviations of Low-Delay coding structure

<img src="network/20200520131610.png" alt="20200520131610" style="zoom: 70%;" />

2. The PSNR standard deviations of Random-Access coding structure

3. Video subjective quality comparison

Test instruction using pre-trained model

We have listed our pre-trained model of EDCNN module, if you want to compare our method, you can substitute the trained EDCNN model in your method. The pre-trained model is placed in weights/edcnn.

python3 predict.py --model [pretrained model] --dir_demo [demo images directory] --save_name [directory to save] --pre_train [weightfile]

Arguments

• n_threads: number of threads for data loading
• cpu: use cpu only
• dir_demo: demo image directory
• model: model name
• pre_train: pretrained model directory
• save_name: directory to save

Citation

Z. Pan, X. Yi, Y. Zhang, B. Jeon and S. Kwong, "Efficient In-Loop Filtering Based on Enhanced Deep Convolutional Neural Networks for HEVC," in IEEE Transactions on Image Processing, vol. 29, pp. 5352-5366, 2020