C4MTS: Challenge on Categorizing Missing Traffic Signs from Contextual Cues
摘要
Traffic signs, despite being crucial for road safety, frequently remain absent. This challenge provides 200 scenes from a recent Missing Traffic Signs Video Dataset (MTSVD), distributed over four types of missing traffic signs: left-hand-curve, right-hand-curve, gap-in-median, and side-road-left, individually observed with their respective contextual cues. 2000 training images, each containing one of the four traffic signs with corresponding bounding boxes, are provided. Two tasks are proposed for the challenge: (i) Object Detection, wherein the model is trained using bounding-box annotations, and (ii) Missing Traffic Sign Scene Categorization, wherein the model is trained using road scene images with in-painted traffic signs, provided with the challenge dataset. Baselines were provided to the participants for both tasks.54 teams registered for the challenge. Overall, the participants could improve the top-1 accuracy significantly by a margin of \(31.5\%\) over the baseline. This work presents the MTSVD in detail, challenge baselines, and the methodology undertaken by the top 2 teams.