yolov3-ios

Using yolo v3 object detection on ios platform.

Example applications:

QuickStart:

Run tiny_model.xcodeproj in ios.

Training

The training process mainly consults qqwweee/keras-yolo3. We add yolov3 with Densnet.

1.Requirement

• python 3.6.4
• keras 2.1.5
• tensorflow 1.6.0

2.Generate datasets

Generate datasets with VOC format. And try python voc_annotations.

3.Start training

• cd yolov3_with_Densenet

For yolo model with darknet:

• wget https://pjreddie.com/media/files/darknet53.conv.74
• rename it as darknet53.weights
• python convert.py -w darknet53.cfg darknet53.weights model_data/darknet53_weights.h5
• python yolov3_train.py, with model_data/darknet53_weights.h5 as pre-trained model

For yolo model with densenet：

• python densenet_train.py, with model_data/dense121_weights.h5 as pre-trained model

Converting

1.Building environment

virtualenv -p /usr/bin/python2.7 keras_coreml_virt
source keras_coreml_virt/bin/activate
pip install protobuf
pip install tensorflow==1.6.0
pip install keras==2.1.5
pip install h5py
pip install coremltools==0.8.0

2.Convert .h5 model to .mlmodel

python coreml.py

Building project in Xcode

• open tiny_model.xcodeproj with Xcode 9+
• change the .mlmodel file and Target Menmbership

For yolo model with darknet or densenet

• modify the code from line 43 to line 49 in YOLO.swift
• change the labels and the anchors in Helpers.swift
• run the project

For tiny model

• just change the labels and run the project