@Article{cmc.2026.085338,
AUTHOR = {Tiyao Zhang, Xue Yuan},
TITLE = {Boundary Measure Alignment via Optimal Transport for Temporal Action Detection},
JOURNAL = {Computers, Materials \& Continua},
VOLUME = {89},
YEAR = {2026},
NUMBER = {2},
PAGES = {--},
URL = {http://www.techscience.com/cmc/v89n2/68788},
ISSN = {1546-2226},
ABSTRACT = {Temporal action detection aims to localize action instances in untrimmed videos and recognize their categories. Although recent detectors have achieved strong performance, accurate boundary localization remains challenging due to gradual action transitions, temporal ambiguity, and annotation uncertainty. Existing boundary supervision usually relies on point-wise classification or local regression losses, which compare predictions and targets at corresponding temporal positions but do not explicitly model temporal displacement between misaligned boundary responses. To address this issue, this paper proposes Boundary Measure Alignment (BMA), a training-stage auxiliary objective for temporal action detection. BMA represents predicted and annotated action starts and ends as probability measures over the feature-level temporal domain. Predicted boundary measures are constructed from the rising and falling edges of the foreground probability trajectory, while target measures are generated by Gaussian smoothing around annotated boundaries. Entropic optimal transport is then used to align predicted and target measures, allowing the loss to explicitly encode temporal boundary displacement. During inference, the BMA branch is removed, so the original detector pipeline remains unchanged. We integrate BMA into AFSD, FCOS, and an ActionFormer-style detector. Experiments on THUMOS14 and ActivityNet-v1.3 show consistent average mAP improvements. On THUMOS14, BMA improves the average mAP of AFSD, FCOS, and the ActionFormer-style detector from 52.0%, 45.3%, and 66.8% to 53.3%, 47.1%, and 69.1%, respectively. On ActivityNet-v1.3, the corresponding average mAP improves from 34.40%, 32.30%, and 35.60% to 34.83%, 32.82%, and 36.04%.},
DOI = {10.32604/cmc.2026.085338}
}