Computer Vision - ACCV 2024 - 17th Asian Conference on Computer Vision, Hanoi, Vietnam, December 8-12, 2024, Proceedings, Part VII

2024-12-24

Asian Conference on Computer Vision

AI Summary

U-Net’s skip connections enhance detail preservation but incur substantial GPU memory overhead, hindering deployment on resource-constrained edge devices. To jointly optimize memory efficiency and representational capacity in lightweight vision models, this work introduces a novel architectural paradigm integrating self-supervised learning, sparse skip-connection reparameterization, and multimodal feature alignment. The framework spans diverse vision tasks—including image understanding, neural radiance field (NeRF)-based 3D reconstruction, cross-domain generalization, and embodied perception. Systematically consolidating over 30 peer-reviewed papers from ACCV 2024 Workshop Session 7, the proposed methods establish new state-of-the-art performance for lightweight models on benchmarks such as Cityscapes and ScanNet, while enabling real-time inference on edge hardware.

Technology Category

• Computer Vision: Multi-modal Vision
• Machine Learning: Multimodal Learning
• Intelligent Robots: Multimodal Perception & Sensor Fusion

Application Category

• Graph Algorithms and Modeling for the Web: Graph embeddings and representation learning for Web-related graphs
• Search and Retrieval-Augmented AI: Retrieval-Augmented Generation (RAG) and multi-modal RAG
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Problem

• Memory Efficiency
• Feature Recognition
• U-Net Optimization

Innovation

• UNet--
• Multi-Scale Information Aggregation Module (MSIAM)
• Information Enhancement Module (IEM)