When HEVC encoder finishes encoding a YUV sequence, it outputs some information on the command line, like this:

According to "Calculation of average PSNR differences between RD-Curves, VCEG-M33" and "Improvements of the BD-PSNR model, VCEG-AI11", Bjontegaard metric is often used to compare two methods. To be specific, Bjøntegaard delta bit rate (BDBR) and Bjøntegaard delta peak signal-to-noise rate (BD-PSNR). We have the output of HEVC encoder, and here are some ready-to-use tools to calculate BDBR and BD-PSNR from the output (which gives us the bitrate and Y-PSNR).

NOTE: These tools are all from the Internet.

The data you need to prepare first before using any of the tools are at least four sets of data like this: (bitrate, Y-PSNR). You need four sets generated from original HEVC encoder and four sets generated from deep-learning-based encoder. Then, you can get the increase in BDBR and the degradation in BD-PSNR.

Read the manual and get the python code in zip-file Bjontegaard-python3.zip.

BD-PSNR/Rate computation tool for five data points, JCTVC-B055

Use the code in zip-file JCTVC-B055.zip.

Code available at GitHub-bjontegaard2