search

maiminh1996 / YOLOv3-tensorflow

Re-implement YOLOv3 with TensorFlow
1 stars 0 forks source link

why the pre_model has no use? LOSS is high #10

Open bubulv opened 6 years ago

bubulv commented 6 years ago

I convert my yolov3_final.weights for my own dataset to yolov3_final.h5, and I use python3 train.py for training. But the loss is also very high. It's the same as I use the darknet53.weights. My yolov3_final.weights is the final train weight for my dataset on darknet, in darknet my cfg and weights are ok and loss is very low. If the yolov3_final.h5 converts right, the training loss should be low too. The convert.py is download from https://github.com/qqwweee/keras-yolo3 Can you help me? Mybe I make some mistake?

asgq123 commented 5 years ago

@bubulv Hello ,I try to train my own data ,but loss is nan , Do you modify the code except data path during training?