In this research we improve the robustness of uncertainty estimation by modeling the loss function using a heavy-tailed distribution and robustness of the loss function is evaluated using the complex regression tasks such as Keypoint estimation.
- A keypoint or feature in an image which can be defined as an unique meaningful structure in that image, however it is semantically ill-defined, in the sense that it is unclear what keypoints are relevant for any given input image.
- More advanced computer vision tasks such as structure from motion, object recognition, 3D reconstruction, simultaneous localization and mapping (SLAM), content-based retrieval and image matching depends on keypoint detection and description methods.
- An unbiased evalution over state-of-the-art keypoint detection and the proposed methods using key-point detection datasets, like face recognition or object detection, is crucial.
- Hence in this research we focus on robustness study for uncertainty estimation for key-point detection task.
- The output of DNNs are stochastic in nature which involves randomness and uncertainties
- Detection of inaccurate measurement leads to increase in decision risks.
- Uncertainty estimation is a challenging problem especially in high dimensional data
- Presence of noisy labels or outliers in the data makes it even more challenging for uncertainty estimation.
Furthermore, this work aims to answer the following research questions.
- Systematic Literature Review on uncertainty estimation methods for Keypoint detection
- What are the different papers in keypoint detection?
- Which are the different tasks solved using keypoint detection?
- What datasets are available for each keypoint detection task?
- What loss functions are used for keypoint detection task?
- Comparison of uncertainty estimation methods for Keypoint detection?
- Robustness study on uncertainty estimation methods for Keypoint detection