Last Friday, my advisor asked me why the time taken by the entropy coding in my method was twice that of another method. My intuitive understanding was that his method optimized the C++ library used for entropy coding, rather than directly using the entropy coding provided by CompressAI.

So yesterday, I spent a whole day adapting my method's entropy coding to his optimized version. To my surprise, the speed of entropy coding turned out to be exactly the same! ＞﹏＜

Today, I spent another full day investigating the cause, analyzing from multiple angles such as whether the bpp (bits per pixel) were similar. The result?! The input data volume for my entropy coding was more than three times that of his!

How absurd! I've wasted two entire days.