Indoor scene recognition using Deep Learning Model
Dataset URL: https://bit.ly/41Fkh1m
Country: USA (MIT university)
15620 images in all 67 Indoor categories (Living Room, Airport)
A minimum of 100 photos in each category
Every image is in the jpg format.
The data set is 2.61 GB.
Data Augmentation: Two transformation pipelines are utilized using PyTorch's transformations module to standardise input images.
The split-folders package is used to ensure an even distribution of each class across Train and test subsets.
Models utilized are: ResNet34, ResNext101_32x16d_wsl, GoogleNet and ShuffleNet.(Minimize risk of vanishing gradient, light weight, state-of-the-art performance , Requires limited computational resources)
Model selection is based on complexity, accuracy, and suitability for the available computational resources and dataset size.
Model accuracy is used as a measure of performance.
