Summary of A-bdd: Leveraging Data Augmentations For Safe Autonomous Driving in Adverse Weather and Lighting, by Felix Assion et al.
A-BDD: Leveraging Data Augmentations for Safe Autonomous Driving in Adverse Weather and Lighting
by Felix Assion, Florens Gressner, Nitin Augustine, Jona Klemenc, Ahmed Hammam, Alexandre Krattinger, Holger Trittenbach, Anja Philippsen, Sascha Riemer
First submitted to arxiv on: 12 Aug 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel machine learning (ML) approach, A-BDD, is proposed to overcome the challenges of perception algorithms in high-autonomy vehicles. Despite impressive performance in fair weather scenarios, ML models are heavily affected by adverse weather and lighting conditions. To address this issue, researchers typically rely on comprehensive real-world datasets, but collecting such data for critical areas of the operational design domain (ODD) can be difficult. As a result, synthetic data is often required for perception training and safety validation. A-BDD is a large set of over 60,000 synthetically augmented images based on BDD100K, with semantic segmentation and bounding box annotations. The dataset includes augmented data for various weather conditions, such as rain, fog, and sunglare/shadow, with varying intensity levels. Novel strategies are introduced to identify useful augmented and real-world data using feature-based image quality metrics like FID and CMMD. Experimental results on A-BDD provide evidence that data augmentations can play a crucial role in closing performance gaps in adverse weather conditions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A-BDD is a new way to help self-driving cars understand their environment better. Right now, these cars are great at seeing things clearly when it’s sunny or not cloudy, but they get confused when the weather is bad. To make them better, researchers need a lot of data about different weather conditions. They can’t always collect this data from real-world cameras because it’s hard to do in certain areas. So, they use computers to create fake images that are similar to what the car would see in different weather conditions. A-BDD is a collection of over 60,000 of these fake images with labels that tell the computer what things are and where they are. The researchers used special tricks to make sure the fake images look like real pictures. They tested their method on A-BDD and found that it helps the cars see better in bad weather. |
Keywords
» Artificial intelligence » Bounding box » Machine learning » Semantic segmentation » Synthetic data
