Abstract: Highlights•We aim to accurately segment faces from product videos for emotion classification.•We propose a segmentation model that can be used on scenic images with occlusions.•We consider a feedback loop that iteratively improves the results of segmentation.
Loading