YOLOv2 - Caffe Inference

This is an implementation of YOLO v2 inference engine using Caffe (pyCaffe) framework. We are not going to train YOLO model from scratch, but convert it using the provided converter instead.

What's inside

• Yolo to Caffe model converter (thanks to Duangenquan)

• Sample application of YOLO v2 on PyCaffe (single image) -- caffe-yolov2/yolo_main.py (upgraded version of yolo main from Xingwangsfu)

• Validation, e.g. ..* subprocess script for running the main script several times ..* mAP calculation using VOC dataset (thanks to AlexeyAB)

Supported models

Tiny YOLO v2 cfg weights

How to use it:

NOTE !!! The following instructions assume that you already have a running Caffe distribution in Python.

Caffe v1.0 Installation instruction here

Single image inference

1. Prepare the config file and pre-trained weights of the model (e.g tiny-yolo-voc)
2. Convert it to Caffe representations(.prototxt and .caffemodel) using the provided script
3. Run the yolo_main.py (add -h for arguments instructions)

Validating the performance

1. As an example, we will be using 2012_val.txt from VOC Dataset as our validation sets.
2. Run caffe_valid_run.py to run our YOLO parser in Caffe, yolo_main.py, against the validation sets.
3. It will produce the needed format in folder results

---

Credits

This application uses Open Source components. You can find the source code of their open source projects below. We acknowledge and are grateful to these developers for their contributions to open source.

Project: Caffe-YOLO by Xingwangsfu (https://github.com/xingwangsfu/caffe-yolo) for the main application of YOLO v1 using pyCaffe

Project: Darknet by AlexeyAB (originally from pjreddie) https://github.com/AlexeyAB/darknet for the mAP calculation on PascalVOC

Project: YoloV2NCS by duangenquan https://github.com/duangenquan/YoloV2NCS for the YOLO v2 output parser and region parameter implementation