A machine learning engineer is developing an object detection model to identify traffic signs. The model achieves high accuracy on the initial, clean training dataset but exhibits poor generalization performance during field tests. Specifically, it fails to correctly identify signs that are partially hidden by foliage and performs inconsistently under adverse weather conditions like rain and sun glare. To improve the model's robustness and real-world performance, which set of data augmentation techniques should the engineer prioritize?
Rotation, scaling, and flipping
Semantic segmentation, object tracking, and sensor fusion
The correct choice is occlusion, spurious noise, and filter application. Occlusion techniques, such as random erasing or cutouts, directly simulate objects being partially blocked, which would help the model recognize signs hidden by foliage. Spurious noise injection mimics visual artifacts like rain or sensor noise. Photometric transformations, such as applying filters to alter brightness, contrast, and color, are effective at simulating varying lighting conditions and glare. The other options are incorrect. Rotation, scaling, and flipping are geometric augmentations that help a model become invariant to changes in orientation and size, but they do not directly address partial occlusion or weather effects. Semantic segmentation, object tracking, and sensor fusion are distinct computer vision tasks or concepts, not data augmentation techniques applied to training images. Lemmatization, n-grams, and stop words are techniques used in Natural Language Processing (NLP) and are not relevant to this computer vision problem.
Ask Bash
Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.
What is occlusion in data augmentation?
Open an interactive chat with Bash
How does spurious noise injection improve model performance?
Open an interactive chat with Bash
What are photometric transformations, and why are they important?