Related Experiment Video
Updated: Dec 11, 2025
08:25
Combining Eye-tracking Data with an Analysis of Video Content from Free-viewing a Video of a Walk in an Urban Park Environment
Published on: May 7, 2019
9.4K
Object Discovery From a Single Unlabeled Image by Mining Frequent Itemset With Multi-scale Features.
Summary
Object Location Mining (OLM) discovers dominant objects in single images using pattern mining and CNN features. This method effectively identifies and localizes objects, outperforming existing techniques in various benchmarks.
Area of Science:
- Computer Vision
- Machine Learning
- Data Mining
Background:
- Object localization in unlabeled images is challenging.
- Existing methods like colocalization and weakly-supervised localization have limitations.
Purpose of the Study:
- To propose a novel method for discovering dominant objects in single, unlabeled images.
- To leverage pattern mining and convolutional neural network (CNN) features for object localization.
Main Methods:
- Developed Object Location Mining (OLM), a pattern mining-based approach.
- Converted CNN feature maps into transactions for pattern discovery.
- Merged discovered patterns (co-occurrence highlighted regions) to identify and localize objects.
Main Results:
- OLM achieves competitive object localization performance against state-of-the-art methods.
- The approach shows strong results compared to unsupervised saliency detection methods across seven benchmarks.
- Accurate localization of entire objects and parts benefits fine-grained classification.
Conclusions:
- OLM is an effective and simple method for unsupervised object discovery and localization.
- The technique demonstrates robustness and broad applicability across diverse datasets.
- The method shows potential for improving downstream tasks like fine-grained image classification.
