Visual Attention Prompted Prediction and Learning
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2023
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866910705718919168 |
|---|---|
| author | Zhang, Yifei Gu, Siyi Pan, Bo Bai, Guangji Qiu, Meikang Yang, Xiaofeng Zhao, Liang |
| author_facet | Zhang, Yifei Gu, Siyi Pan, Bo Bai, Guangji Qiu, Meikang Yang, Xiaofeng Zhao, Liang |
| contents | Visual explanation (attention)-guided learning uses not only labels but also explanations to guide model reasoning process. While visual attention-guided learning has shown promising results, it requires a large number of explanation annotations that are time-consuming to prepare. However, in many real-world situations, it is usually desired to prompt the model with visual attention without model retraining. For example, when doing AI-assisted cancer classification on a medical image, users (e.g., clinicians) can provide the AI model with visual attention prompt on which areas are indispensable and which are precluded. Despite its promising objectives, achieving visual attention-prompted prediction presents several major challenges: 1) How can the visual prompt be effectively integrated into the model's reasoning process? 2) How should the model handle samples that lack visual prompts? 3) What is the impact on the model's performance when a visual prompt is imperfect? This paper introduces a novel framework for attention-prompted prediction and learning, utilizing visual prompts to steer the model's reasoning process. To improve performance in non-prompted situations and align it with prompted scenarios, we propose a co-training approach for both non-prompted and prompted models, ensuring they share similar parameters and activations. Additionally, for instances where the visual prompt does not encompass the entire input image, we have developed innovative attention prompt refinement methods. These methods interpolate the incomplete prompts while maintaining alignment with the model's explanations. Extensive experiments on four datasets demonstrate the effectiveness of our proposed framework in enhancing predictions for samples both with and without prompt. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2310_08420 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | Visual Attention Prompted Prediction and Learning Zhang, Yifei Gu, Siyi Pan, Bo Bai, Guangji Qiu, Meikang Yang, Xiaofeng Zhao, Liang Computer Vision and Pattern Recognition Visual explanation (attention)-guided learning uses not only labels but also explanations to guide model reasoning process. While visual attention-guided learning has shown promising results, it requires a large number of explanation annotations that are time-consuming to prepare. However, in many real-world situations, it is usually desired to prompt the model with visual attention without model retraining. For example, when doing AI-assisted cancer classification on a medical image, users (e.g., clinicians) can provide the AI model with visual attention prompt on which areas are indispensable and which are precluded. Despite its promising objectives, achieving visual attention-prompted prediction presents several major challenges: 1) How can the visual prompt be effectively integrated into the model's reasoning process? 2) How should the model handle samples that lack visual prompts? 3) What is the impact on the model's performance when a visual prompt is imperfect? This paper introduces a novel framework for attention-prompted prediction and learning, utilizing visual prompts to steer the model's reasoning process. To improve performance in non-prompted situations and align it with prompted scenarios, we propose a co-training approach for both non-prompted and prompted models, ensuring they share similar parameters and activations. Additionally, for instances where the visual prompt does not encompass the entire input image, we have developed innovative attention prompt refinement methods. These methods interpolate the incomplete prompts while maintaining alignment with the model's explanations. Extensive experiments on four datasets demonstrate the effectiveness of our proposed framework in enhancing predictions for samples both with and without prompt. |
| title | Visual Attention Prompted Prediction and Learning |
| topic | Computer Vision and Pattern Recognition |
| url | https://arxiv.org/abs/2310.08420 |