| Group |
2021-14 |
Status |
completed |
| Title |
SmoothMix: a system for creating new images from existing ones |
| Supervisor |
Maria Amer |
| Description |
Deep learning is getting a lot of attention because it achieves unprecedented levels of accuracy, which in some cases exceeds human performance. Deep learning, however, requires large amount of data for training. Data augmentation aims to produce inexpensive new data from the original data. It has been proven effective not only to enhance the accuracy of deep neural network but also leads to a better generalization, even with limited dataset.
In this project, the students will implement a system that creates new data (images) from original data by smoothly blending two images to create a new image. The students are required to test different blending methods and then to integrate their system in a deep learning application of their liking: object detection, object tracking, image classification, human pose estimation, person identification, or activity recognition. Often the codes of deep learning applications are available publicly (such as YOLOv3 https://pjreddie.com/darknet/yolo/ for object detection or Mask-RCNN https://github.com/facebookresearch/detectron2 for human pose estimation) and needs to be integrate with the system the students will develop.
This project (creating new data from existing ones and integrating into available application) is a complex multi-disciplinary design problem. The project involves communications, control systems, signal processing, computer vision, artificial intelligence, and software design.
|
| Requirement |
Signals and Systems, Signal processing, Image processing, Programming (C/C++, Python, Matlab) |
| Tools |
Linux, OpenCV, Deep learning packages (such as Tensorflow and Pytorch). |
| Number of Students |
6 |
| Students |
Karim Rhoualem, Ahmed Ali, Tianyang Jin, Paulina Navarro Aviles, Shuang Luo, Alexander Wolfe |
| Comments: |
|
| Links: |
|