Vol. 12, Issue 2 (2023)

Author(s):
Ambreen Hamadani, Nazir A Ganai, Ayush Rajan Gupta, Faizan Farooq Shah, Mohd Shakeb Umar, Kartik Jindal, Saurabh Jajoo, Harsh Gopalika and J Bashir

Abstract:
An attempt was made to train several deep convolutional neural networks to classify images from the Image Detect challenge into 200 distinct classes. Multiple CNN architectures were tried which included pretrained modes with feature extraction as well as development of models from scratch. Our models consisted of 3- 5 convolutional layers, 3 max-pooling layers, 2-3 fully connected layers, and a SoftMax classification layer. Among the multiple transfer learning techniques used, Inception Res Net V2 showed the most promising results with the highest accuracy and lowest loss. The test predictions were 56% accurate and our model was among the top 13 models among 39 submissions (top 36%) in the Image classification task.