A simple Moon tracker based on mask-RCNN.
The whole evolution of moon-tracker can be found in the below link.
Customized by : Praveen Vijayan
Inspired from : https://towardsdatascience.com/object-detection-using-mask-r-cnn-on-a-custom-dataset-4f79ab692f6d
Inspired from : https://github.com/matterport/Mask_RCNN
Follow the below steps to create any custom object mask and object locator of your interest.
- Use Google's Colab if you don't have a GPU to work with. Same code will work there easily. Also it will reduce the error based on the tensorflow dependency.
Keep your files in Google drive and mount the same in the colab environment. And change the working directory to the moon-tracker(in drive) and now executing the code will be a cake walk. (Also you can upload to the runtime session without mounting to the Drive).
https://colab.research.google.com/notebooks/io.ipynb
Also the moon-tracker code will work in normal PC with CPU environment.
-
Tensorflow and keras may have some version issues. As some the function calls from '/matterport/Mask_RCNN' is of the old version. Create your environment accordingly else you have to modify inside the actual functions. Recommended to use tensorflow 1.4+ and Keras 2.0.8+.
-
Use labellmg to create a mask for an image. It will give output as an XML file.
- Download and keep the COCO weights from 'mask_rcnn_coco.h5' in the master.
Link : https://github.com/matterport/Mask_RCNN/releases
-
Play with learning_rate, epoch and layers ('all' , '3+', '4+' , 'heads' ) for better accuracy.
-
Notebook by default saves the trained model in moon_model. If an improved model is used while testing it can result in better accuracy.
-
Now explore moon_mask.ipynb to run your moon-tracker.
