artamaral
commented
5 years ago Try for COCO: Change coco.names and leave only "person", and change the number of classes to 1 on coco.data.
Open akde opened 5 years ago
artamaral
commented
5 years ago Try for COCO: Change coco.names and leave only "person", and change the number of classes to 1 on coco.data.
akde
commented
5 years ago @artamaral thx for the immediate reply.
changed coco.names into coco_hmn.names so only "person" remains. changed coco.data into coco_hmn.data so number of classes is 1 now.
executed following command: ./darknet detector test cfg/coco_hmn.data cfg/yolov3-tiny.cfg yolov3-tiny.weights data/dog.jpg
here is the output: dog: 57% car: 52% truck 56% car: 62% truck: 43% bicycle: 59%
should I change the cfg file too?
bit-scientist
commented
5 years ago @akde merhaba abi! Did you happen to find .weights and .cfg fiels for humans only? If yes, could you share them? Thanks
Dear all and @AlexeyAB
I have to run yolov3-tiny on a very constrained system in order to detect humans only. Since I do NOT have a system with GPU, I cannot train them. Finding everything in the scene then displaying humans only is NOT a desired solution since I want to improve FPS as much as possible.
So is there a pre-trained yolov3-tiny weights for only humans? Where can I get such a pre-trained weights? It would be awesome if someone could share it.
Looking forward for your reply...