#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.
- Instead of taking a single image file path, the modified darknet CLI now takes a file that contains a list of image file paths to predict objects for, one image file path per line.
- Instead of writing out an image annotated with bounding boxes, the modified darknet CLI writes out a TSV file of bounding box annotations, one per input image file. The TSV file has the same path and basename as the input image and a .tsv suffix. Format of the file is (image file path, predicted entity, prediction probability, top left x, top left y, bottom right x, bottom right y).
The old command to run the pre-trained YOLO detector was as follows:
./darknet detector test cfg/coco.data cfg/yolo.cfg yolo.weights data/dog.jpg
Post the changes described above, the command is:
./darknet detector test cfg/coco.data cfg/yolo.cfg yolo.weights mylist.txt
where mylist.txt looks like this:
data/dog.jpg
data/person.jpg
Example output for bounding boxes for dog.jpg looks like this:
data/dog.jpg car 0.542268 247 56 364 125
data/dog.jpg truck 0.261111 247 56 364 125
data/dog.jpg bicycle 0.509336 53 89 301 317
data/dog.jpg cat 0.346415 63 152 173 390
data/dog.jpg dog 0.558766 63 152 173 390