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
2.07k stars 501 forks source link

Random Crop and Ignore Small Faces #59

Open WANG-MengJiao opened 3 years ago

WANG-MengJiao commented 3 years ago

Hi, there. The paper indicates that augmentation method like 'Ignore Small Faces' and 'Random Crop' can boost the performance. However, I didn't find these augmentations in the source codes. Is these methods implemented in this repository?

wjtan99 commented 3 years ago

Look at utils/face_datasets.py line 333.
The ignoring small faces is in data preparation.

alicera commented 2 years ago

Here is the your test result https://github.com/deepcam-cn/yolov5-face/blob/master/data/images/result.jpg

Here is the yolov5l-face.py detect result (https://drive.google.com/file/d/16F-3AjdQBn9p3nMhStUxfDNAE_1bOF_r/edit) image

The small face are not detected. Is it the reason for ignoring small faces is in data preparation?

Tallisgo commented 2 years ago

Look at utils/face_datasets.py line 333. The ignoring small faces is in data preparation.

@wjtan99 hello, can you share the value of ignore_small?