Top 3 submission at IITB Techfest'21 by Mihir Karandikar, Tanishq Selot, Atharva Mohite and Yeeshukant singh (Students at IIT Indore)
Dataset Preparation:
-
The script Generate Masks(1024x1024).py will generate the masks from the given annotations. The masks generated have dimensions of 1024x1024
-
Due to memory issues in Collab, the mask size has been reduced to 512x512 by appropriately reducing the image size and at the same time not affecting the proportions adversly.
Training :
- The python notebook VBL_UNET_Training contains all the code for training of the model and also evaluation on the dataseet
Generation of Segmented Images
- The script 'VBL_UNET_Predictions.py' uses the saved model to generate the segmented images which are then saved in the folder VBL_Segmented_Images