Summary of Proto-ood: Enhancing Ood Object Detection with Prototype Feature Similarity, by Junkun Chen et al.
Proto-OOD: Enhancing OOD Object Detection with Prototype Feature Similarity
by Junkun Chen, Jilin Mei, Liang Chen, Fangzhou Zhao, Yan Xing, Yu Hu
First submitted to arxiv on: 9 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
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 The proposed Proto-OOD network structure is designed to address the issue of neural networks mispredicting out-of-distribution objects when trained on limited category samples. By leveraging prototype features in few-shot learning, Proto-OOD enhances the representativeness of category prototypes using contrastive loss and detects OOD data by evaluating the similarity between input features and category prototypes. The network generates OOD samples for training the similarity module with a negative embedding generator during training. Experimental results on Pascal VOC and MS-COCO datasets show that Proto-OOD significantly reduces the false positive rate, outperforming existing methods. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about how to make sure neural networks don’t get confused when they see new things. When these networks are only trained on a few examples of something, they often have trouble recognizing things outside of those examples. The researchers propose a new way to fix this problem called Proto-OOD. It works by creating a set of “prototypes” that represent what the network has learned about each category. Then, it compares the input data to these prototypes to decide if it’s part of the expected category or not. This approach is shown to be very effective in reducing mistakes. |
Keywords
» Artificial intelligence » Contrastive loss » Embedding » Few shot
