Reproduce “XiDepth: a Lightweight and Efficient Network for Self-supervised Monocular Depth Estimation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Elena Izzo, Riccardo Toniolo, Lamberto Ballan
Researchers from the University of Padova developed XiDepth, a lightweight and energy-efficient network for self-supervised monocular depth estimation targeting embedded devices. The model achieves competitive depth accuracy on the KITTI dataset with only 0.8 million parameters and demonstrates 35% lower inference time and energy consumption compared to leading methods when evaluated on a Raspberry Pi 4.
Key insights
- The energy-efficient XiNet block, originally designed for image classification, can be effectively adapted for dense prediction tasks like monocular depth estimation, extending its proven utility.
- Direct empirical evaluation on embedded hardware reveals that avoiding specific operations (e.g., depth-wise convolutions, attention mechanisms) can yield substantial real-world improvements in energy efficiency and inference speed.
- A carefully designed lightweight architecture, combined with self-supervised learning, can maintain competitive depth estimation accuracy while drastically reducing model size and computational demands.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.