Have darknet output machine readable (json) predictions instead of writing the results directly to the bounding boxes of an image.
For this purpose darknet detector now recognizes the -tagger-output switch, which will disable all writes to stdout except when the predictions are written.
Example:
./darknet detector test cfg/coco.data cfg/yolo.cfg yolo.weights data/dog.jpg -tagger-output 2>/dev/null
{"input": "data/dog.jpg", "time": 13.036422, "matches": [
{"class": "dog", "probability": 0.823520, "left": 132, "right": 321, "top": 231, "bottom": 521 }
, {"class": "truck", "probability": 0.643006, "left": 467, "right": 680, "top": 84, "bottom": 168 }
, {"class": "bicycle", "probability": 0.852180, "left": 95, "right": 588, "top": 123, "bottom": 448 }
], "count": 3 }
#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.