I noticed the existing discussion in #1141 about the interpolation implementation in the tracking evaluator, but my question is about the evaluation interpretation rather than the implementation. Since online trackers are expected to output results frame by frame without access to future observations, could you clarify how this post-hoc tracklet interpolation should be interpreted, and whether it may partially compensate for short-term missed detections before computing the final tracking metrics?
I noticed the existing discussion in #1141 about the interpolation implementation in the tracking evaluator, but my question is about the evaluation interpretation rather than the implementation. Since online trackers are expected to output results frame by frame without access to future observations, could you clarify how this post-hoc tracklet interpolation should be interpreted, and whether it may partially compensate for short-term missed detections before computing the final tracking metrics?