September 2021
tl;dr: Multi-sensor expansion of previous work on corner cases.
An essential task of ML-based perception models are to reliably detect and interpret unusual new and potentially dangerous situations. A complication for the development of corner case detectors is the lack of consistent definitions, terms and corner case descriptions.
Corner cases introduced not only by vision system, but also radar/lidar and sensor fusion. This is an extension from CC-vision and CC-vision (IV version).
- Summaries of the key ideas
- Object level corner cases of lidar: dust or smoke cloud
- Questions and notes on how to improve/revise the current work