tommyjtl / google-colab-tinyyolov2-model-training-script

Automated Model Training Script on Google CoLab

Run following script/command one by one

• python setup.py
• python generate_conf.py
  • Use python generate_conf_v2.py if you want the training process stop automatically when avg loss is lower than 0.06
• cd ./conf/ && bash start_train.sh && cd ..
  • Use cd ./conf/ && python train.py && cd .. if you used generate_conf_v2.py for generating configuration files
• cd ./conf/ && bash test_train.sh && cd ..
• python export_model.py
• python copy_to_googledrive.py

User input:

• Please enter your dataset directory location on Google Drive:
  • must have directories for each class
  • must have a .names file that includes the names of each class that you want to recognized
• Please enter the number of classes you would like to train:
• Please enter the name of your project:

Upload your labeled dataset to Google Drive

Put all your image data to on folder (remember to categorize images of each class to their designated folder), the file structure will look something like this:

- DATASET_NAME
	- CLASS_1_NAME
		- CLASS_1_0.jpg
		- CLASS_1_0.txt
		- CLASS_1_1.jpg
		- CLASS_1_1.txt
		- CLASS_1_2.jpg
		- CLASS_1_2.txt
		...
		- CLASS_1_n.jpg
		- CLASS_1_n.txt
	- CLASS_2_NAME
		- CLASS_2_0.jpg
		- CLASS_2_0.txt
		- CLASS_2_1.jpg
		- CLASS_2_1.txt
		- CLASS_2_2.jpg
		- CLASS_2_2.txt
		- CLASS_2_n.jpg
		- CLASS_2_n.txt
		...
	...
	- CLASS_N_NAME
		- CLASS_N_0.jpg
		- CLASS_N_0.txt
		- CLASS_N_1.jpg
		- CLASS_N_1.txt
		- CLASS_N_2.jpg
		- CLASS_N_2.txt
		- CLASS_N_n.jpg
		- CLASS_N_n.txt
		...

• Warning: Before you do the labeling work, please do not use such an image that its storage size is larger than 1MB, you can use photo processing software or online tool to resize it.

Reference:

• When should I stop training
• How to train YOLOv2 to detect custom objects