Some discussion points on Section 4.7: Qualitative Results

Hi, I noticed that the bird's legs moving rapidly were not segmented into the mask. Could this be due to inconsistencies between the training datasets of EdgeTAM and SAM2? For example, differences in motion patterns, annotation quality for leg regions, or data distribution. Could the authors clarify whether training data discrepancies might affect performance on fast-moving or detailed regions? Thank you!