CVPR Findings2026
FALCON: Fast Adaptive Lightweight Computation of Intensities and Events for Depth Estimation
Sankarshana Venugopal, Mohammad Mostafavi, Jonghyun Choi
摘要
Event-based stereo depth estimation promises high-speed, robust perception under challenging conditions, but fusing events with intensity frames incurs heavy computational costs, hindering real-time deployment. We introduce FAL-CON: a Fast Adaptive Lightweight Computation framework that co-designs spiking neuron dynamics (via LIF neurons) with 1×1 convolutional event representations and a streamlined stereo backbone for efficient depth estimation. Our per-pixel spiking encoders capture temporal dynamics with over 1000× fewer FLOPs than prior modules, while our σ-matching heuristic optimizes event grouping to minimize blur. FALCON achieves state-of-the-art accuracy on DSEC (MAE 0.356) and MVSEC (MDE 11.7 cm) at 32.2 FPS and 72.8 FPS, respectively, establishing a new real-time benchmark for neuromorphic vision applications. Our code is publicly available at https://github.com/snumprlab/falcon.
问 Lune
几分钟读透这篇 Paper,不用再花一下午。
Lune 能读完全文,也能读引用它的顶会 Paper,讲清它的结论、证据和局限,每一点都注明出自哪篇 Paper。不用硬啃,也不靠猜。
脉络
- Stereo Depth from Events Cameras: Concentrate and Focus on the Future 引用了 Event-Intensity Stereo: Estimating Depth by the Best of Both Worlds
- Discrete time convolution for fast event-based stereo 引用了 Event-Intensity Stereo: Estimating Depth by the Best of Both Worlds
- Discrete time convolution for fast event-based stereo 引用了 Deep Event Stereo Leveraged by Event-to-Image Translation
它引用的 Paper8
- Mind the Gap: Understanding the Modality Gap in Multi-modal Contrastive Representation LearningWeixin Liang, Yuhui Zhang, Yongchan Kwon, Serena Yeung 等NeurIPS 2022 · 834 次引用
- Stereo Depth from Events Cameras: Concentrate and Focus on the FutureYeongwoo Nam, S. Mohammad Mostafavi I., Kuk-Jin Yoon, Jonghyun ChoiCVPR 2022 · 56 次引用
- Event-Intensity Stereo: Estimating Depth by the Best of Both WorldsS. Mohammad Mostafavi I., Kuk-Jin Yoon, Jonghyun ChoiICCV 2021 · 45 次引用
- Deep Event Stereo Leveraged by Event-to-Image TranslationSoikat Hasan Ahmed, Hae Woong Jang, S. M. Nadim Uddin, Yong Ju JungAAAI 2021 · 41 次引用
- Discrete time convolution for fast event-based stereoKaixuan Zhang, Kaiwei Che, Jianguo Zhang, Jie Cheng 等CVPR 2022 · 34 次引用
相关 Paper
- SpikeStereoNet: A Brain-Inspired Framework for Stereo Depth Estimation from Spike StreamsZhuoheng Gao, Yihao Li, Jiyao Zhang, Rui Zhao 等ICLR 2026 · 2 次引用
- DERD-Net: Learning Depth from Event-based Ray DensitiesDiego de Oliveira Hitzges, Suman Ghosh, Guillermo GallegoNeurIPS 2025 · 6 次引用
- Active Event-based Stereo VisionJianing Li, Yunjian Zhang, Haiqian Han, Xiangyang JiCVPR 2025
- CREST: An Efficient Conjointly-trained Spike-driven Framework for Event-based Object Detection Exploiting Spatiotemporal DynamicsRuixin Mao, Aoyu Shen, Lin Tang, Jun ZhouAAAI 2025
- 240FPS Stereo Vision from Monocular Mixed SpikesYeliduosi Xiaokaiti, Yakun Chang, Yang Bai, Zhaojun Huang 等CVPR 2026
