Tracking Any Object Amodally

• DOI: 10.48550/arxiv.2312.12433
• Publication Date: 2023-01-01
• Authors: Cheng-Yen Hsieh; Tarasha Khurana; Achal Dave; Deva Ramanan
• Keywords: Tracking (education)

Amodal perception, the ability to comprehend complete object structures from partial visibility, is a fundamental skill, even for infants. Its significance extends applications like autonomous driving, where clear understanding of heavily occluded objects essential. However, modern detection and tracking algorithms often overlook this critical capability, perhaps due prevalence \textit{modal} annotations in most benchmarks. To address scarcity amodal benchmarks, we introduce TAO-Amodal, featuring 833 diverse categories thousands video sequences. Our dataset includes \textit{amodal} modal bounding boxes visible partially or fully objects, including those that are out camera frame. We investigate current lay land both by benchmarking state-of-the-art trackers segmentation methods. find existing methods, when adapted tracking, struggle detect track under heavy occlusion. mitigate this, explore simple finetuning schemes can increase metrics 2.1\% 3.3\%.