Daily Arxiv

This is a page that curates AI-related papers published worldwide.
All content here is summarized using Google Gemini and operated on a non-profit basis.
Copyright for each paper belongs to the authors and their institutions; please make sure to credit the source when sharing.

SpiderNets: Estimating Fear Ratings of Spider-Related Images with Vision Models

Created by
  • Haebom

Author

Dominik Pegler, David Steyrl, Mengfan Zhang, Alexander Karner, Jozsef Arato, Frank Scharnowski, Filip Melinscak

Outline

This paper investigates whether pre-trained computer vision models can accurately predict fear levels in spider-related images, providing a foundational study for developing an adaptive computer exposure therapy system. Using transfer learning, three different models were applied to predict human fear ratings (on a 0-100 scale) from a standardized dataset of 313 images. Cross-validation results showed a mean absolute error (MAE) of 10.1–11.0. Learning curve analysis revealed that reducing the dataset size resulted in a decrease in performance, but further increasing the dataset size did not significantly improve performance. Explainability assessment demonstrated that the model's predictions were based on spider-related features, and category-specific error analysis identified visual conditions associated with high error rates, such as distant views and artificial/painted spiders. This study demonstrates the potential of explainable computer vision models for fear rating prediction and highlights the importance of model explainability and sufficient dataset size for the development of effective emotion-recognition therapy.

Takeaways, Limitations

•
Takeaways:
◦
We demonstrate that a pre-trained computer vision model can be used to predict fear levels.
◦
It presents the potential to contribute to the development of adaptive computer exposure therapy systems.
◦
Emphasizes the importance of ensuring the model's explainability.
◦
Confirm the need for sufficient dataset size.
•
Limitations:
◦
Relatively high error rate of MAE 10.1~11.0.
◦
Possible degradation of generalization performance due to limitations in dataset size.
◦
High error rates under certain visual conditions (distant views, artificial/painted spiders, etc.).
◦
Because the study was limited to spider phobia, generalizability to other phobias is limited.
👍
Made with Slashpage