henrymxu / darknet

Convolutional Neural Networks

Custom Darknet

This version of darknet allows the user to save the results of classifications. The results are saved as a json file containing the various detections along with their properties (class, bounding box).

To run this custom command, use validate.

Single Image Validation

Running the detector on a single image located at ~/Documents/ImageSet/gallardo/image_39.jpg and outputting the results at ~/Documents/ImageSet/gallardo/results.json.

./darknet validate cfg/yolov3.cfg cfg/yolov3.weights ~/Documents/ImageSet/gallardo/result.json ~/Documents/ImageSet/gallardo/image_39.jpg

• (Assuming you are in the darknet directory)

Multiple Image Validation

To run the detector on multiple images, leave the image path blank. Once the weights are loaded, input each image path one at a time. The results will still all be saved in the same file. Input a blank file name to end the process.

./darknet validate cfg/yolov3.cfg cfg/yolov3.weights ~/Documents/ImageSet/gallardo/result.json

• (Assuming you are in the darknet directory)

The images that are passed in cannot use ~ as apart of the path.

To automate the usage of the detector with multiple images, create a text file with all the paths of the images you wish to be validated and run the following command.

cat image_list.txt | ./darknet detect cfg/yolov3.cfg yolov3.weights ~/Documents/ImageSet/gallardo/result.json

• (Assuming you are in the darknet directory)
• (Assuming image path list is found in image_list.txt with paths seperated by new lines)

Darknet

Darknet is an open source neural network framework written in C and CUDA. It is fast, easy to install, and supports CPU and GPU computation.

For more information see the Darknet project website.

For questions or issues please use the Google Group.