Surf makesense for annotating your pictures.
-
try to grab your boxes a little bigger than the objects.
-
export annotation as textes.
if you want to train yolov3 on google colab you don't need to download cuda, cudnn and opencv.
training yolov3 on google colab --> YOLOV3-COLAB
surf this link for building OpenCV GPU
surf this link for building Darknet.
Try to run your weight in CPP
My GPU : 1050Ti
10 FPS in python
25 FPS in cpp
Warning: change the weight and cfg configuration to CUDA (change the CPU and OPENCV to CUDA)