search

david8862 / keras-YOLOv3-model-set

end-to-end YOLOv4/v3/v2 object detection pipeline, implemented on tf.keras with different technologies
MIT License
639 stars 221 forks source link

yolov4 train problems #46

Open asd834 opened 4 years ago

asd834 commented 4 years ago

image image

why?

david8862 commented 4 years ago

seems you're training on windows environment. this repo has not officially support windows, but I'll try to add some fix to see if it could fit for you.

david8862 commented 4 years ago

@asd834 I've fixed some path issue to improve the compatibility for windows env. You can try if it could work now

asd834 commented 4 years ago

@david8862 image val_loss is too high tensorflow-gpu Training time is too long

david8862 commented 4 years ago

seems your dataset is a bit too small (totally 1000 images) for training a big detection model to make sure convergence is correct. Regarding the training speed, not sure about your CUDA env but seems the GPU is not active.