Open Tomonobu3110 opened 4 years ago
@Tomonobu3110 the backbone for YOLOv2-Tiny is based on the Darknet Reference Model, and you can download pretrained weights for it at https://pjreddie.com. If you want pretrained weights for the entire network, then you can get them here.
@LlewellynS96 Thanks for the information. I have one question.
Config file of the Darknet Reference Model says "width = 256, height = 256". But config file of YOLOv2-tiny is "width = 416, height = 416".
It seems the network is different between the Darknet Reference Model and Yolov2-tiny. Can I use the weight file of Darknet Reference Model for YOLOv2-tiny training?
BR, Saito
Is there anyone who knows how can we get the Pretrained Convolutional Weights for yolov2-tiny ?
I'm using darknet + yolov2 on Raspberry Pi3. But it takes about 6 min to finish the detection per a picture (about 600 x 400). To speed up this, I try to use yolov2-tiny but accuracy is lower than I expected. I decided to train yolov2-tiny by coco data (with reducing the vocabulary). I prepared data (jpg and txt) for training and validation. But I couldn't find "Pretrained Convolutional Weights" for yolov2-tiny.
Maybe this is for yolov2 (not for yolov2-tiny). https://pjreddie.com/media/files/darknet19_448.conv.23
BR, Saito