Eric Liu
Eric Liu
I think the batch size = 1 can't not be split to multi-gpu training in test phase , so you can close the test phase and start training .
Unfortunately , I don't have multi-gpu computer or environment :( So , it is really hard for me , maybe you can see this issue https://github.com/eric612/MobileNet-YOLO/issues/28
Refer this issue https://github.com/eric612/MobileNet-YOLO/issues/198
@guagua11 Remove https://github.com/eric612/MobileNet-YOLO/blob/master/models/mobilenetv2_voc/yolo_lite/solver.prototxt#L2-L4
This is a trick to increase mAP , which be used in [alexeyAB project](https://github.com/AlexeyAB/darknet/blob/master/src/yolo_layer.c#L516-#L557) , but I think it may produce some false bbox in real case . The first...
Yes , I usually use GIOU here , I forgot to check MSE term .
The fast way is wait AlexeyAB implement and copy his code . 😄 https://github.com/AlexeyAB/darknet/issues/4147
Update , You can reference this project https://github.com/jwchoi384/Gaussian_YOLOv3
@hitzht , You can reference my prototxt , but I think it have some training problem , I always got a nan loss when IOU closed to gt [test.txt](https://github.com/eric612/MobileNet-YOLO/files/3877887/test.txt) [train.txt](https://github.com/eric612/MobileNet-YOLO/files/3877888/train.txt)...
You are right.