YOLO-v2: Real-Time Object Detection

Still under development. 71 mAP(darknet) and 74mAP(resnet50) on VOC2007 achieved so far.

This is a pre-released version.

What's new

This repo is now deprecated, I am migrating to the latest Gluon-CV which is more user friendly and has a lot more algorithms in development.

• Pretrained YOLOv3 models which achiveve 81%+ mAP on VOC and near 37% mAP on COCO: Model Zoo.

• Object Detection model tutorials.

This repo will not receive active development, however, you can continue use it with the mxnet 1.1.0(probably 1.2.0).

Disclaimer

This is a re-implementation of original yolo v2 which is based on darknet. The arXiv paper is available here.

Demo

Getting started

• Build from source, this is required because this example is not merged, some custom operators are not presented in official MXNet. Instructions
• Install required packages: cv2, matplotlib

Try the demo

• Download the pretrained model(darknet as backbone), or this model(resnet50 as backbone) and extract to model/ directory.
• Run

# cd /path/to/mxnet-yolo
python demo.py --cpu
# available options
python demo.py -h

Train the model

• Grab a pretrained model, e.g. darknet19
• (optional) Grab a pretrained resnet50 model, resnet-50-0000.params,resnet-50-symbol.json, this will produce slightly better mAP than darknet in my experiments.
• Download PASCAL VOC dataset.

cd /path/to/where_you_store_datasets/
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtrainval_06-Nov-2007.tar
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2007/VOCtest_06-Nov-2007.tar
# Extract the data.
tar -xvf VOCtrainval_11-May-2012.tar
tar -xvf VOCtrainval_06-Nov-2007.tar
tar -xvf VOCtest_06-Nov-2007.tar
ln -s /path/to/VOCdevkit /path/to/mxnet-yolo/data/VOCdevkit

• Create packed binary file for faster training

# cd /path/to/mxnet-ssd
bash tools/prepare_pascal.sh
# or if you are using windows
python tools/prepare_dataset.py --dataset pascal --year 2007,2012 --set trainval --target ./data/train.lst
python tools/prepare_dataset.py --dataset pascal --year 2007 --set test --target ./data/val.lst --shuffle False

• Start training

python train.py --gpus 0,1,2,3 --epoch 0
# choose different networks, such as resnet50_yolo
python train.py --gpus 0,1,2,3 --network resnet50_yolo --data-shape 416 --pretrained model/resnet-50 --epoch 0
# see advanced arguments for training
python train.py -h