Garden-Nerd-Flower-Recognition

Image Processing competition hosted on HackerEarth

This is a multi-classification problem with 102 types of flowers.

Approach

• Used Pre-Trained networks like ResNets and DenseNets.
• Trained first on 64X64 sized images, followed by 128X128, followed by 224X224 sized images. ( Progressive Resizing )
• Used Image Augmentations like Cutout, Horizontal and Vertical Flips etc.
• Used techniques like MixUp
• Ensembled top 10 performing models.

F-1 Score - 89.48 % on the Public LeaderBoard