search

deepcam-cn / yolov5-face

YOLO5Face: Why Reinventing a Face Detector (https://arxiv.org/abs/2105.12931) ECCV Workshops 2022)
GNU General Public License v3.0
2.11k stars 502 forks source link

about some detals #17

Closed www516717402 closed 3 years ago

www516717402 commented 3 years ago

Thank you for open this repo. I have some questions as follows:

  1. All model use 800 image size for input in readme.md table ? However 640 images size in evaluate ? https://github.com/deepcam-cn/yolov5-face/blob/f4db424d7066ca30b32d81cac634ec846e2b0c5f/train.py#L439

  2. Have you mean filter small face by this code? Anything else?

    As we explain before, the Mosaic has to work with the ignoring small faces, otherwise the performance degrades dramatically

https://github.com/deepcam-cn/yolov5-face/blob/f4db424d7066ca30b32d81cac634ec846e2b0c5f/utils/datasets.py#L900

derronqi commented 3 years ago

1、p5 800, p6 832, widerface test 640 2、no, check the train2val.py

www516717402 commented 3 years ago

@derronqi Thank you for reply. About second question, doesn't use any method to filter small face in your code. However use this method to incress 3.% mAP of Easy datset in your paper.

https://github.com/deepcam-cn/yolov5-face/blob/2320920838813fc4513fbc9b49acc160f05e73be/data/train2yolo.py#L129 https://github.com/deepcam-cn/yolov5-face/blob/2320920838813fc4513fbc9b49acc160f05e73be/data/val2yolo.py#L29