help wanted
仓库指标
- Star
- (3,830 star)
- PR 合并指标
- (30 天内没有已合并 PR)
描述
As highlighted in #10, the current MTCNN design is appropriate for the project while a single detection method is implemented, but with multiple, it makes more sense to separate the return of bounding boxes and confidences from the cropping of images.
Should also think about returning facial landmarks to implement face alignment following detection.
Retinaface is probably the best option to start with.