yolo2_light

Light version of convolutional neural network Yolo v2 for objects detection with a minimum of dependencies (INT8-inference, BIT1-XNOR-inference)

This repository supports:

Windows - cpu only

How to compile:

Build yolo_cpu.sln

How to use INT8-inference:

Use flag -quantized at the end of command, for example,
- yolo_cpu.exe detector test tiny-yolov2/voc.names tiny-yolov2/yolov2-tiny-voc.cfg tiny-yolov2/yolov2-tiny-voc.weights -thresh 0.10 person.jpg -quantized
For the custom dataset, you should use input_calibration= parameter in your cfg-file, from the correspon cfg-file: yolov2-tiny-voc.cfg, ...

How to use BIT1-XNOR-inference:

You should base your cfg-file on tiny-yolo-obj_xnor.cfg and train it by using this repository as usual https://github.com/AlexeyAB/darknet
Then use it for Detection-test:
- yolo_cpu.exe detector test tiny-yolo-voc_xnor/voc.names tiny-yolo-voc_xnor/tiny_yolo_xnor.cfg tiny-yolo-voc_xnor/tiny_yolo_xnor.weights -thresh 0.21 person.jpg
- Note: here we provide tiny_yolo_xnor.weights which was trained in voc dataset with tiny_yolo_xnor-train.cfg