Summary of A Realistic Protocol For Evaluation Of Weakly Supervised Object Localization, by Shakeeb Murtaza et al.
A Realistic Protocol for Evaluation of Weakly Supervised Object Localization
by Shakeeb Murtaza, Soufiane Belharbi, Marco Pedersoli, Eric Granger
First submitted to arxiv on: 15 Apr 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 The paper proposes a new evaluation protocol for Weakly Supervised Object Localization (WSOL) that eliminates the need for manual bounding box annotations. Currently, WSOL methods rely on validation sets with bounding box annotations for model selection and test sets with bounding box annotations for threshold estimation. However, this approach is not suitable for real-world scenarios where these annotations are unavailable. The authors demonstrate a significant decline in localization performance when using only class labels and image features for model selection and threshold estimation. To address this issue, they generate noisy pseudo-boxes from off-the-shelf region proposal methods like Selective Search, CLIP, and RPN for model selection and threshold estimation. Experimental results on ILSVRC and CUB datasets show that the proposed protocol facilitates better model selection and threshold estimation, leading to comparable localization performance with ground-truth bounding boxes. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary The paper solves a problem in deep learning by making it easier to train models without manual annotations. Currently, training models for object detection requires thousands of hours of work just to label the data. The authors find that this process is not necessary and can be replaced by using existing methods to generate good-enough bounding boxes. This makes it possible to train better models faster and more accurately. |
Keywords
» Artificial intelligence » Bounding box » Deep learning » Object detection » Region proposal » Supervised
