RetinaDetection是基于RetinaFace修改的检测方法,原论文is a practical single-stage SOTA face detector which is initially described in arXiv technical report
- Organise the dataset directory as follows:
data/retinaface/
train/
images/
label.txt
val/
images/
label.txt
test/
images/
label.txt
- Install MXNet with GPU support.
- Install Deformable Convolution V2 operator from Deformable-ConvNets if you use the DCN based backbone.
- Type
make
to build cxx tools.
Please check train.py
for training.
- Copy
rcnn/sample_config.py
torcnn/config.py
为了获得更好的训练效果,可针对性的修改一些参数,如下:
config.TRAIN.MIN_BOX_SIZE = 5 #最小bbox
config.FACE_LANDMARK = False #使用landmark
config.USE_BLUR = False
config.BBOX_MASK_THRESH = 0
config.COLOR_MODE = 2 #增强
config.COLOR_JITTERING = 0.125
-
Download pretrained models and put them into
model/
.ImageNet ResNet50 (baidu cloud and dropbox).
ImageNet ResNet152 (baidu cloud and dropbox).
-
Start training with
sh train_model.sh
.
Before training, you can check theresnet
network configuration (e.g. pretrained model path, anchor setting and learning rate policy etc..) inrcnn/config.py
.
Please check test.py
for testing.
释放训练好的模型
@inproceedings{yangsai1991@163.com,
year={2019}
}