Summary of Learnable Prompt For Few-shot Semantic Segmentation in Remote Sensing Domain, by Steve Andreas Immanuel et al.
Learnable Prompt for Few-Shot Semantic Segmentation in Remote Sensing Domain
by Steve Andreas Immanuel, Hagai Raja Sinulingga
First submitted to arxiv on: 16 Apr 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 This paper addresses the challenge of few-shot segmentation, where a model must segment novel object classes within an image based on only a few annotated examples. The authors propose a method to mitigate catastrophic forgetting, which occurs when adding novel classes affects the performance on base classes. They use SegGPT as their base model and train it on the base classes before applying separate learnable prompts for each novel class. To handle varying object sizes in remote sensing images, they employ patch-based prediction and a patch-and-stitch technique to address discontinuities along patch boundaries. The authors also utilize image similarity search and prompt selection during inference to reduce false positive predictions. Their proposed method improves the weighted mIoU on the OpenEarthMap dataset from 15.96 to 35.08. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper is about a way for computers to quickly learn to identify new objects in pictures, even if they’ve only seen a few examples of those objects before. The problem is that when you add new objects to the picture, it can forget how to identify the old ones. To solve this, the authors use a special type of computer model called SegGPT and train it on the old objects first. Then, they give it special instructions for each new object. This helps the computer remember how to identify both old and new objects. The authors also found a way to make the computer better at handling different sizes of objects in pictures, which is important for remote sensing applications like monitoring the environment from space. |
Keywords
» Artificial intelligence » Few shot » Inference » Prompt
