search

david8862 / keras-YOLOv3-model-set

end-to-end YOLOv4/v3/v2 object detection pipeline, implemented on tf.keras with different technologies
MIT License
638 stars 220 forks source link

yolov4 training time #101

Open amindehnavi opened 4 years ago

amindehnavi commented 4 years ago

Hi David, when I use Kitti dataset to train "yolov4_darknet" or "yolov4_efficientnet", each epoch takes about 4h to complete! could you help me to fix that? I'm using colab and its gpu

annotation and class files: train (1).txt class.txt

robisen1 commented 4 years ago

That may or may not be unreasonable. What is your GPU? Are the images just from the Kiti dataset? Have you changed any hyperparameters

amindehnavi commented 4 years ago

I have no idea what its gpu type is. In colab i changed "runtimetype" to gpu . About dataset, yes all images are from kitti.also I use default hyperparameters and change nothing

david8862 commented 4 years ago

@amindehnavi didn't meet such issue before. One possible problem is your image file are all PNG format. Not sure if it has any impact on loading & decoding.