# Neural Video Coding using Specialization _by Henry Félix (B<>Com & Orange) - 2022.10.13_ ###### tags: `VAADER` `Seminar` ![](https://i.imgur.com/FWn55R4.png) ## Abstract We study the effect of overfitting a neural video codec to the sequence being encoded. This allows a higher compression ratio. The new parameters of the decoder are transmitted differentially with respect to a universal neural decoder, using an existing lossy compression standard, whose quantization noise is taken into account during the overfitting. Using a test set of 4K video sequences, we show that this leads to an average gain in compression of 15.8% over all bitrates. At the higher bitrates typically used for this type of content, the decoder overhead is proportionally smaller, and the average gain over all test sequences is 21.9%.