This document proposes two scalable point cloud (PC) geometry codecs, submitted to the JPEG Call for Evidence on Point Cloud Coding (PCC) , notably targeting two different types of scalability:
1) Resolution Scalable Deep Learning-based Point Cloud Geometry Coding (RS-DLPCC) : This codec provides scalability on the number of points, called here resolution scalability; since the scalable layers are independently coded, this codec also offers multiple description coding, i.e. all layers by themselves offer useful PC reconstructions, which is a very interesting feature for error resilience, e.g. limiting the effect of packet losses in specific layers, etc.
2) Quality Scalable Deep Learning-based Point Cloud Geometry Coding (QS-DLPCC) : This codec provides quality scalability using layer dependent coding, meaning that decoding a layer requires decoding the previous layers as well, thus not offering multiple description coding.
The proposed scalable codecs are based on recent developments in deep learning-based PC geometry coding (ADL-PCC) , and offer the key functionalities targeted by the JPEG Call for Evidence, notably number of points or resolution scalability, quality scalability, and spatial random access.
The proposed RS-DLPCC and QS-DLPCC coding solutions offer a compression efficiency that is rather competitive with the MPEG G-PCC standard , whereas the non-scalable version (ADL-PCC) of the proposed codecs is able to achieve significant RD performance gains over the G-PCC standard. Nevertheless, since these are some of the first (if not the first) deep learning-based scalable geometry coding solutions in the literature, the proposed scalable codecs shall be regarded more as a proof of concept as it is clear that substantial performance improvements may be expected in the future.