corlov
commented
5 years ago I launched the darknet program and got confused. I want to recognize headlights on a car. I have found the corresponding class in 9k.names file. I have posted my results below, where do I need to look for a problem?
Weight file and data file I got from there: http://pjreddie.com/media/files/yolo9000.weights https://github.com/pjreddie/darknet/blob/master/cfg/combine9k.data
./darknet detector test cfg/combine9k.data cfg/yolo9000.cfg yolo9000.weights data/helicopter.jpg
....
mask_scale: Using default '1.000000'
Loading weights from yolo9000.weights...Done!
data/helicopter.jpg: Predicted in 21.284352 seconds.
The copter not found.
./darknet detector test cfg/combine9k.data cfg/yolo9000.cfg yolo9000.weights data/vesta.jpg
...
mask_scale: Using default '1.000000'
Loading weights from yolo9000.weights...Done!
data/vesta.jpg: Predicted in 20.433591 seconds.
sedan: 67%
The headlights not found.
./darknet detector test cfg/combine9k.data cfg/yolo9000.cfg yolo9000.weights data/car.jpg
...
mask_scale: Using default '1.000000'
Loading weights from yolo9000.weights...Done!
data/car.jpg: Predicted in 20.775150 seconds.
The car and headlights not found.
./darknet detector test cfg/combine9k.data cfg/yolo9000.cfg yolo9000.weights data/woman.jpg
...
mask_scale: Using default '1.000000'
Loading weights from yolo9000.weights...Done!
data/woman.jpg: Predicted in 20.228838 seconds.
The woman not found.My input images are:
car.jpg
helicopter.jpg
vesta.jpg
woman.jpg
I see into the work of YOLOv3. Then I run detection with yolov3.cfg , yolov3.weights and coco.names (80 class list) all working good at video and images, and accuracy too fine. But if I change cfg to yolov3-openimages.cfg weights to yolov3-openimages.weights and openimages.name(601 class list) accuracy is so bad. At the images all detections are common. Bee like animal, horse like enimal and many other objects not recognized. Can someone advise me what to do?