Object recognition algorithms have become more robust in last decade with recent advances in domain and with competitions such as ImageNet which created more focus on the area among researchers and engineers. While a variety of approaches have been used, object recognition task has few common procedures. Representation of images and category learning and detection differ these variety of approaches. In this project, algorithms for visual recognition will be reviewed, implemented and compared with each other on given objective.
I would like to express my gratitude to the project’s principal investigator Prof. Dr. Bülent Sankur and also Dr. Erdem Yörük, my research supervisors, for their patient guidance, encouragement and useful critiques of this research work through the semester.
- Data cleaning by hand may needed due to possible anomalies in dataset.
- Re-implementation of BoVW models on new dataset with ablation study and application of SVM with intersection kernel as classifier
- Overall review of works of first semester
- Comparison of Image classification with convolutional neural networks
- Overview of transfer learning methods and outcomes of fine-tuning approaches as represented in Figure 5 on the dataset
- Processing time comparison of image classification models
- Solving scale problem in between classes ( Coca Cola 1 Liter & Coca Cola 1.5 Liter)
