SEDD-PCC

A Single Encoder-Dual Decoder Framework For End-To-End Learned Point Cloud Compression

A Single Encoder-Dual Decoder Framework For End-To-End Learned Point Cloud Compression

Kai-Hsiang Hsieh (Author) · Monyneath Yim (Co-Author) · Jui-Chiu Chiang (Adviser)

National Chung Cheng University, TaiwanNational Chung Cheng University, Taiwan

Abstract

Abstract

Three-stage architecture

Fig. 1 Existing Learning-based point cloud compression solutions, (a) Geometry compression. (b) Attribute compression.
(c) Joint geometry and attribute compression through recoloring.
(d) Proposed SEDD-PCC: an end-to-end learned point cloud compression scheme.

圖 1 現有基於深度學習之點雲壓縮方法,(a) 幾何壓縮。(b) 屬性壓縮。
(c) 透過重新上色的聯合幾何和屬性壓縮。
(d) 提出的 SEDD-PCC:聯合幾何與屬性之點雲壓縮方法。

🔍 Problem: Most learning-based compression schemes treat geometry and attribute coding separately, employing distinct encoders and decoders. This not only increases computational complexity but also fails to fully exploit shared features between geometry and attributes.

🔍 問題: 大多數基於深度學習的壓縮方法將幾何和屬性分開處理,使用不同的編碼器和解碼器。這不僅增加了計算複雜性,而且無法充分利用幾何和屬性之間的共享特徵。

✅ Our Solution: To address this limitation, we propose SEDD-PCC, an end-to-end learning-based framework for lossy point cloud compression that jointly compresses geometry and attributes. SEDD-PCC employs a single encoder to extract shared geometric and attribute features into a unified latent space, followed by dual specialized decoders that sequentially reconstruct geometry and attributes.

✅ 我們的解決方法: 為了解決這一限制,我們提出了 SEDD-PCC,這是一個端到端的基於深度學習的有損點雲壓縮架構,可以聯合壓縮幾何和屬性。SEDD-PCC 採用單一編碼器將共享的幾何和屬性特徵提取到統一的潛在空間中,然後透過兩個量身打造解碼器首先重建幾何資訊再將其資訊傳遞給屬性解碼器來完成整個重建過程。

🎯 Enhancement: We incorporate knowledge distillation to enhance feature representation learning from a teacher model, further improving coding efficiency. With its simple yet effective design, SEDD-PCC provides an efficient and practical solution for point cloud compression.

🎯 增強: 我們結合知識蒸餾來增強從教師模型中學習特徵,進一步提高編碼效率。憑藉其簡單而有效的設計,SEDD-PCC為點雲壓縮提供了高效且實用的解決方法。

📊 Results: Comparative evaluations against both rule-based and learning-based methods demonstrate its competitive performance, highlighting SEDD-PCC as a promising AI-driven compression approach.

📊 結果: 與傳統的MPEG標準相比基於深度學習的點雲壓縮方法證明了其具有競爭力的性能,凸顯了SEDD-PCC作為一種有前途的 AI 驅動壓縮方法。

Contributions

Contributions

🔹 Unified Architecture:
The first unified model that employs a single encoder with two specialized decoders for joint compression of point cloud geometry and attributes, leveraging a simple yet effective autoencoder architecture. 首個採用單一編碼器和兩個專用解碼器的統一模型,用於點雲幾何和屬性的聯合壓縮,利用簡單而有效的自動編碼器架構。
🔹 Shared Latent Space:
SEDD-PCC jointly compresses geometry and attributes using a single encoder to extract shared features into a unified latent space, followed by dual decoders to sequentially reconstruct both components. SEDD-PCC 使用單一編碼器將共享特徵提取到統一的潛在空間中來聯合壓縮幾何和屬性,然後通過雙解碼器按順序重建幾何與屬性。
🔹 Lightweight & Effective:
The framework achieves competitive performance with other learning-based methods while maintaining a lightweight and efficient design, highlighting its potential as a robust solution for AI-driven point cloud compression. 該框架與其他基於深度學習的方法相比具有競爭力的性能,同時保持輕量級和高效的設計,凸顯其作為 AI 驅動點雲壓縮的強大解決方法的潛力。

Method

Method

Three-stage architecture

Figure 2: The three-stage training strategy.

圖 2:三階段訓練策略。

Stage 1: Attribute compression ; Stage 2: Geometry compression, adopting a teacher model for knowledge distillation ;
Stage 3: Fine-tune all components by initializing Encoder_U and Decoder_A from Stage 1, and Decoder_G from Stage 2.
第一階段: 屬性壓縮;第二階段: 幾何壓縮,使用教師模型進行知識蒸餾;
第三階段: 通過初始化第一階段的 Encoder_U 和 Decoder_A,以及第二階段的 Decoder_G 來微調整體架構。

Network

Network

Three-stage architecture

Figure 3: The proposed SEDD-PCC architecture.

圖 3:提出的 SEDD-PCC 架構。

The proposed SEDD-PCC architecture, illustrated in Fig. 3, consists of a single encoder and dual decoders designed for joint compressing of geometry and attributes. 圖 3 所示的提出的 SEDD-PCC 架構,由單一編碼器和雙解碼器組成,旨在聯合壓縮幾何和屬性。

Experiment results

Experiment results

Fig 1 Fig 2 Fig 3 Fig 4 Fig 5 Fig 6

Fig. 4 R-D performance of the proposed scheme in terms of 1-PCQM.

圖 4 根據 1-PCQM 指標提出方法的 R-D 性能。

Table 1. BD-rate (%) against G-PCCv23 for various schemes.

表 1. 各種方法對比 G-PCCv23 的 BD-rate (%)。

BD-rate comparison

BibTeX Citation

BibTeX Citation

@inproceedings{hsieh2025seddpcc, title={SEDD-PCC: A Single Encoder--Dual Decoder Framework for End-to-End Learned Point Cloud Compression}, author={Kai-Hsiang Hseh, Monyneath Yim and Jui-Chiu Chiang}, year={2025} }