Contributors: Tushar Sawant | Ashwin Sahsrabuddhe | Self.
The project makes use of Convolutional Neural Network to detect certain Indoor Scenes from the MIT Places 365 Dataset. We used only 3 categories - Bedroom, Corridor and Kitchen for purpose of this project. We also created our custom dataset by downloading images from Google Images for testing purposes.
The repository contains 2 folders:
-
CNN from scratch: This folder contains code for the CNN model implemented from scratch which gives accuracy of about 77.67% .
-
Transfer Learning: This folder contains code for CNN model developed using transfer learning. Some layers of AlexNet were used for transfer learning. This gave accuracy of around 91.33% .