- Demos for object detection, mask segmentation and keypoints recognition
- YOLO v2 (RegionLossLayer) and v3 (YoloLossLayer) are supported
- Instance Mask segmentation with Yolo
- Keypoints Recognition with yolo
- training data preparation and training
# clone
git clone https://github.com/leon-liangwu/MaskYolo_Caffe.git --recursive
# install requirements
cd ROOT_MaskYolo
pip install -r requirements.txt
# compile box_utils
cd lib/box_utils
python setup.py build_ext --inplace
# compile caffe
cd caffe-maskyolo
cp Makefile.config.example Makefile.config
make -j
make pycaffe
Click here to download pretrained models
cd ROOT_MaskYolo
tar zxvf /your/downlaod/model/path/models.tgz ./
support to use yolo v2 or v3 to detect objects in images
cd tools
python yolo_inference.py [--img_path=xxx.jpg] [--model=xxx.prototxt] [--weights=xxx.caffemodel]
# Net forward time consumed: 3.96ms
The demo result is shown below.
cd ROOT_MaskYolo
sh ./scripts/convert_detection.sh #generate lmdb for detection
cd ./models/mobilenetv2-yolo/
nohup sh yolo_train.sh > train.log &
tail -f train.log
cd tools
python mask_inference.py [--img_path=xxx.jpg] [--model=xxx.prototxt] [--weights=xxx.caffemodel]
# Net forward time consumed: 8.67ms
python kps_inference.py [--img_path=xxx.jpg] [--model=xxx.prototxt] [--weights=xxx.caffemodel]
# Net forward time consumed: 5.58ms
some resulting samples are show below.
I just feed yolo results to roi_pooing
or roi_alignment
layer.
coming soon
You Only Look Once: Unified, Real-Time Object detection http://arxiv.org/abs/1506.02640
YOLO9000: Better, Faster, Stronger https://arxiv.org/abs/1612.08242
YOLOv3: An Incremental Improvement https://pjreddie.com/media/files/papers/YOLOv3.pdf
Faster R-CNN: Towards Real-Time Object Detection with Region Proposal Networks
Mask R-CNN