Resume training custom object detection model in google colab

I followed the all steps in “https://medium.com/deepquestai/train-object-detection-ai-with-6-lines-of-code-6d087063f6ff” to create the custom object detection model. It works well for small set of images and be able to complete training within 12 hours.

However, when I increase a number of labels and a number of training images. The training process is highly likely to take more than 12 hours. As you know, google colab will delete everything after 12 hours.

My question is: how can I resume training model from the saved model with the same state from the last checkpoint? For example, I set 100 Epochs and the latest completed epoch before 12 hours is 78th. How can I resume continue training model by using the saved model from Epoch 78th?

1 Like

@hammer22z Thanks for questions.

You can use your 78th model as the pre-trained for the next training instead of using the pre-trained-yolov3.h5. However, please note that the detection_config.json file of the previous training will no longer be compatible with new models of the saved in the new training process, as a new detection_config.json file will be generated.

BTW: The next version of ImageAI will allow you continue training with previous models and use both new and old models with the same detection_config.json file.