Prathima Krishna Subramanian

- To fine tune a pre-trained object detection Tensorflow API model for labeling images quickly - To perform effective data augmentation to increase the accuracy of prediction results - To implement human in the loop technique to improve the accuracy of prediction - Comparison between annotations provided by CNN based object detectors and manual annotations

- Applying various image transformation techniques to extract directional information from noisy images - Comparing the various techniques to identify the best transformation technique which extracts majority of the noise - Designing a CNN to predicts and classify images containing noise into 10 classes corresponding to 10 noise levels - Optimizing the designed network for maximum accuracy and minimum loss