Nick Green
Mentor: Dr. Rasha Gargees
Research: Leveraging RGB-D Features with Transfer Learning for Image Classification
In this research we examine the image classification metrics when utilizing transfer learning in extracting RGB and depth features. This approach leverages the benefits of high performance pretrained models via transfer learning techniques, allowing for efficient lower-level feature distinction that is further refined on the dataset specific features. The proposed combined model should distinguish key RGB and depth features, enhancing the means to recognize its fed images. This experimentation uses a Lidar and Visual Walking Terrain Dataset, consisting of paired RGB and Depth images of 9 different terrain classes, and compares performances when utilizing both features against two other models that use either feature individually. Lastly, this research delves into concatenation strategies, exploring the potency of transfer learning in leveraging multi-modal data for improved classification performance.