The dataset about ceymo HOT 5 CLOSED

Hi, could you please attach the script you are using for the conversion?

from ceymo.

Thanks, I got the Imagedata by "saving as" with labelme, but I have another problem, how many epochs to train this dataset on Mask-R-CNN to reach the best state, as I finished training before 20 epochs, I am confused about it. @oshadajay

from ceymo.

Cool. We used the Tensorflow Object Detection API and we did not trained the model for a fixed number of epochs. We observed the evaluation metrics which are published from time to time and stopped training when the metrics sort of saturated. Hope this helps.

from ceymo.

Thanks, my loss function was fitted at 20 epoch, I think the reason for this is related to the low diversity of the dataset, after all, many images taken at adjacent times in the dataset have similarities, so the model learns the features of the dataset very quickly.

from ceymo.

Yeah that could be the reason.

from ceymo.