Digital Image Processing - A Unified framework for object detection and semantic segmentation.

Object detection and semantic segmentation are two strongly correlated tasks, yet typically solved separately or sequentially with substantially different techniques. From our brief research of the topic, complementary effects from the typical failure cases of the two tasks were observed, even in state-of-the-art methods. Motivated by this, we implemented a unified framework for joint object detection and semantic segmentation, inspired by the works of J. Dong et al, 2014. By enforcing the consistency between final detection and segmentation results, the implemented unied framework can effectively leverage the advantages of leading techniques for these two tasks.Furthermore, both local and global context information are integrated into the framework to better distinguish the ambiguous samples. All models are jointly trained such that a better decision can be made by the classifier using the information aggregated till the final stage. Extensive experiments on the PASCAL VOC 2012 demonstrate encouraging performance of the proposed unified framework for the semantic segmentation task.

For more information, check our Report