Open inspect_custom_data.ipynb in google colab by providing github repo link.
Select GPU from Edit -> Notebook Settings
Do copy to drive action.
Steps to train model
Run this in a separate code block after making sure GPU is enabled in EDIT->Notebook Settings & you have the initial steps like deleting folder, cloning repo & making sure a couple of images from the dataset are viewable with mask, before proceeding with training(taking 50 mins at the moment on GPU)
After training we will save the generated h5 file to our local machine & use it for inference using CPU. We can also verify the same with another notebook, where we need to upload the h5 file & update the filename in that notebook as well.
To download the h5 weights file, do the following-
Keep updating the same code block with below lines & repeat individually,
%cd ..
%cd logs/
%cd {into latest table, check the training log for h5 file name & folder it is enclosed in.}
#then run the below,
from google.colab import files
files.download('mask_rcnn_table_0010.h5')
Saving weight file on google drive and download
Mounting google drive
from google.colab import drive
drive.mount('/content/gdrive')
Copy the file to google drive %cp mask_rcnn_table_0010.h5 /content/gdrive/'My Drive'/