The goal of this project is to develop a method for image classiffcation and object localization.The method will use a framework that is similar to the R-CNN (region-based convolutional neural network) model proposed by Girshick et al. as discussed to the following papers:
- R. Girshick, J. Donahue, T. Darrell, J. Malik, \Rich Feature Hierarchies for Accurate Object Detection and Semantic Segmentation," IEEE Conference on Computer Vision and Pattern Recognition, 580-587, June 23-28, 2014.
- R. Girshick, J. Donahue, T. Darrell, J. Malik, \Region-Based Convolutional Networks for Accurate Object Detection and Segmentation," IEEE Transactions on Pattern Analysis and Machine Intelligence, 38(1):142-158, January 2016.
We have used a ResNet-50 network that was pre-trained on the ImageNet data set to extract visual features. We have trained a 2-layer feed-forward neural network classifier. Selective Search region proposal algorithm is used to predict the bounding boxes of targets in test images. Quantitative performance evaluation is done in two stages: - Computing the confusion matrix, precision and recall for each object type, and the overall accuracy in terms of the percentage of correctly classified test images
- Computing the percentage of correctly classified and localized test images.