search

deep-diver / Soccer-Ball-Detection-YOLOv2

YOLOv2 trained against custom dataset
116 stars 43 forks source link

Changing checkpoint variable to custom value #21

Open getcontrol opened 4 years ago

getcontrol commented 4 years ago

I have Darflow and your Jupyter notebook working great together. Thanks for this great resource!

I am working off an old Macbook Air with limited CPU/GPU so it trains very slowly. I've gotten up to batch 500 so far and then stopped the kernel to feed the checkpoint into the proceeding steps.

I understand that the lines:

options = {"model": "cfg/yolo_custom.cfg",
           "load": -1,
           "gpu": 1.0}

will utilize my last checkpoint (i.e. 500). Confirmed with output :

Loading from ./ckpt/yolo_custom-500

How do I configure the variable "load": -1, to jump to the provided pretrained weight i.e. 2300?

Thanks again for your great documentation !

Tanmay-Kulkarni101 commented 4 years ago

Hey @getcontrol,

All the checkpoints should be present in the ckpt folder. So, the moment, darkflow manages to save the checkpoint after finishing 2300 steps you will be able to access it. Did this answer your question?