📝 Full Publications List

EDRIC: Embracing 1D Autoencoder for Real-Time Lossy LiDAR Reflectance Compression    Jiahao Zhu, Kang You, Kequan Mao, Dandan Ding*, Zhan Ma

• we introduce EDRIC, a highly effective neural compression framework that offers state-of-the-art compression efficiency while achieving real-time capability. To overcome the suboptimal downsampling scheme and inefficient feature extraction in conventional 3D frameworks, EDRIC serializes a 3D point cloud into a 1D sequence and introduces a lightweight 1D autoencoder to efficiently compress the serialized LiDAR reflectance signal. In addition, we explicitly incorporate geometric priors through a geometry-aware entropy model, effectively exploiting the interdependencies between reflectance attributes and underlying geometry.

Adaptive AV2 In-loop Filtering via Guided Neural Model with Vectorized Quantization    Jiahao Zhu, Kang You, Dandan Ding*, Zhan Ma

• we propose PACE, a new framework that reformulates ancestral context aggregation as a non-causal backbone and confines causality to a lightweight, stage-scalable predictor, eliminating repetitive backbone executions and reducing computational overhead. The predictor supports an arbitrary number of prediction stages, enabling seamless adaptation across diverse performance-latency trade-offs without reloading parameters.

RATE-DISTORTION OPTIMIZED LORA FOR EFFICIENT POST-FILTERING IN AV2    Kequan Mao, Xin Yang, Dandan Ding*, Urvang Joshi, Debargha Mukherjee

• we proposes RD-LORA, a rate-distortion (R-D) optimized low-rank adaptation (LoRA) framework for neural post-filtering in the upcoming AV2 coding standard. RD LoRA adapts a pre-trained base neural model to diverse input content via online updates to LoRA parameters, whose quantity is determined by the matrix ranks. The updated parameters are then quantized, transmitted to the decoder, and merged with the pre-trained weights for post-filtering.

Adaptive AV2 In-loop Filtering via Guided Neural Model with Vectorized Quantization    Kequan Mao, Dandan Ding*, Urvang Joshi, Debargha Mukherjee

• we proposes a neural self-guided filter for AV2, introducing two key techniques: (i) adaptive selection of the number of guidance channels for higher coding efficiency, and (ii) a vector quantization method for lower bitrate overhead. Our method is general and applicable to any network architecture.

Improving Occupancy Prediction for Multiscale Point Cloud Geometry Compression
Zehong Li, Jiahao Zhu, Dandan Ding*, Zhan Ma

• We propose two new techniques. The first is KPA (Key Point-driven Attention), which integrates both local and global characteristics. The second is AdaScale (Adaptive Lossy/Lossless Scale), which decides whether the transitional scale should be in lossless or lossy mode based on temporal displacement, thereby enhancing the reconstruction quality of the temporal reference.

Revisit Point Cloud Quality Assessment: Current Advances and a Multiscale-Inspired Approach
Junzhe Zhang, Tong Chen, Dandan Ding*, Zhan Ma

• We paper proposes PQI, a simple yet efficient metric to index point cloud quality. PQI suggests using scale-wise key points to uniformly perceive distortions within a point cloud, along with a mild neighborhood size associated with each key point for compromised N2N computation.

Scalable Point Cloud Attribute Compression
Junzhe Zhang, Jianqiang Wang, Dandan Ding*, Zhan Ma

• We develops a Scalable Point Cloud Attribute Compression solution, termed ScalablePCAC. In a two-layer example, ScalablePCAC uses the standard G-PCC at the base layer to directly encode the thumbnail point cloud that is downscaled from the original input, and a learning-based model at the enhancement layer to compress and restore the full-resolutioninput point cloud conditioned on the base layer reconstruction.

GeoQE: Enhancing Quality of Experience in Point Cloud Streaming
Junzhe Zhang, Chengfeng Han, Dandan Ding*, Zhan Ma

• we propose GeoQE, an enhancement model that seamlessly integrates with the G-PCC decoder to mitigate compression artifacts and improve QoE.

Efficient LiDAR Reflectance Compression via Scanning Serialization
Jiahao Zhu, Kang You, Dandan Ding*, Zhan Ma

• We introduce SerLiC, a serialization-based neural compression frame work to fully exploit the intrinsic characteristics of LiDAR reflectance.

Reno: Real-time Neural Compression for 3D LiDAR Point Clouds
Kang You, Tong Chen, Dandan Ding, M.Salman Asif, Zhan Ma

• We proposes RENO, the first real-time neural codec for 3D LiDAR point clouds, achieving superior performance with a lightweight model.

Another way to the top: exploit contextual clustering in learned image coding
Yichi Zhang, Zhihao Duan, Ming Lu, Dandan Ding, Fengqing Zhu, Zhan Ma

• We propose Contextual Clustering based LIC (CLIC), which relies on clustering operations and local attention instead of traditional convolutions to generate compact representations for image compression.

Neural Adaptive Loop Filtering for Video Coding: Exploring Multi-Hypothesis Sample Refinement
Dandan Ding, Junjie Wang, Guangkun Zhen, Debargha Mukherjee, Urvang Joshi, Zhan Ma

• We reformulate ALF as a Multi-Hypothesis Sample Refinement (MSR) problem, using a DNN model to generate multiple distortion hypotheses that are linearly superimposed to approximate the final reconstruction.

Sparse Tensor-based Multiscale Representation for Point Cloud Geometry Compression
Jianqiang Wang, Dandan Ding, Zhu Li, Xiaoxing Feng, Chuntong Cao, Zhan Ma

• We propose SparsePCGC, a low-complexity multiscale representation that performs sparse convolutions only on most-probable positively-occupied voxels to characterize spatial correlations efficiently.

Biprediction-Based Video Quality Enhancement via Learning
Dandan Ding, Wenyu Wang, Junchao Tong, Xinbo Gao, Zoe Liu, Yong Fang

• We develop a biprediction-based multiframe video enhancement (PMVE) framework that synthesizes virtual frames to extract cross-correlations between successive frames for high-accuracy quality restoration.

Neural Reference Synthesis for Inter Frame Coding
Dandan Ding, Xiang Gao, Chenran Tang, Zhan Ma

• We propose a Neural Reference Synthesis (NRS) framework with joint optimization of reconstruction enhancement and reference synthesis modules to improve both in-ring filtering and inter-frame prediction.

Point Cloud Upsampling via Perturbation Learning
Dandan Ding, Chi Qiu, Fuchang Liu, Zhigeng Pan

• We propose learning 2D perturbations through MLPs to estimate coordinate shifts from sparse input points to upsampled dense points, outperforming state-of-the-art methods in geometric uniformity.