Skip to content

Latest commit

 

History

History
19 lines (10 loc) · 819 Bytes

File metadata and controls

19 lines (10 loc) · 819 Bytes

Indoor-Scene-Recognition-w-MIT-indoors-dataset

Purpose of this project was to train a image classifier to recognize indoor scene from 67 different category. Classifier cnn modal's accuracy found to be 0.86 for testing. Trained dataset includes 15617 files from all 67 classes. Splitted as 0.7 for train, 0.2 for validation, 0.1 for testing.

allIndoors

MIT Indoor Dataset

https://web.mit.edu/torralba/www/indoor.html

Modal's Acc Loss:

image

TO DO:

-Fine tuning w/ Densenet, VGG16, Alexnet compare the results.

-Try LoRA (Low rank adaptation).