Deep Reinforcement Learning for Active Human Pose Estimation

Gärtner, Erik; Pirinen, Aleksis; Sminchisescu, Cristian

doi:10.1609/aaai.v34i07.6714

“…Social experience depicts the impact of family and friends, colleagues, neighbours, peer groups and even communities in social media [ 109 ]. As AI-enabled devices used in medical services and healthcare can outperform human physicians in making precise measurements to detect diseases [ 110 ], customers will more likely have a positive experience. Thus, accuracy and precision of AI technology enhance user experience, confidence and understanding [ 10 ].…”

Section: Literature Reviewmentioning

confidence: 99%

Applied Artificial Intelligence and user satisfaction: Smartwatch usage for healthcare in Bangladesh during COVID-19

Uzir

¹

,

Halbusi

²

,

Lim

³

et al. 2021

Technology in Society

View full text Add to dashboard Cite

The evolution of Artificial Intelligence (AI) has revolutionized many aspects of human life, including healthcare. Amidst the Covid-19 pandemic, AI-enabled smartwatches are being used to help users to self-monitor and self-manage their health. Using a framework based on Stimulus-Organism-Response (S–O-R) theory, this present study aimed to explore the use of AI-enabled smartwatches for health purposes, in particular the effects of product quality, service quality, perceived convenience, and perceived ease of use on user experience, trust and user satisfaction. Based on a purposive survey sample of 486 smartphone users in Bangladesh, data collected was analyzed using SPSS software for elementary analyses and PLS-SEM for hypotheses testing. The findings showed that the predictors, namely product quality, service quality, perceived convenience, and perceived ease of use, significantly affected user experience and trust. Similarly, user experience and trust were influential on user satisfaction and played partial mediating roles between predictors and user satisfaction. Besides, gender and age moderate the relationships of experience and trust with customer satisfaction. These findings support the S–O-R theoretical framework and have practical implications for brand and marketing managers of smartwatches in developing product features and understanding users' attitudes and behaviours.

show abstract

“…Some of the earliest work [1] provides a general framework for this problem in low resource vision systems and camera control [5]. Recent work in this domain aims at learning view selection strategies to solve diverse tasks including object recognition [29,2], segmentation [8,28], visual navigation [39,10,44], and pose estimation [16,37]. Similar work to ours falls in the domain of active image understanding, in the subsequent subsection we provide a brief literature survey of the constituent modules of our model.…”

Section: Related Workmentioning

confidence: 98%

Glimpse-Attend-and-Explore: Self-Attention for Active Visual Exploration

Seifi¹,

Pacák²,

Tuytelaars³

2021

Preprint

0

View full text Add to dashboard Cite

Active visual exploration aims to assist an agent with a limited field of view to understand its environment based on partial observations made by choosing the best viewing directions in the scene. Recent methods have tried to address this problem either by using reinforcement learning, which is difficult to train, or by uncertainty maps, which are task-specific and can only be implemented for dense prediction tasks. In this paper, we propose the Glimpse-Attendand-Explore model which: (a) employs self-attention to guide the visual exploration instead of task-specific uncertainty maps; (b) can be used for both dense and sparse prediction tasks; and (c) uses a contrastive stream to further improve the representations learned. Unlike previous works, we show the application of our model on multiple tasks like reconstruction, segmentation and classification. Our model provides encouraging results while being less dependent on dataset bias in driving the exploration. We further perform an ablation study to investigate the features and attention learned by our model. Finally, we show that our self-attention module learns to attend different regions of the scene by minimizing the loss on the downstream task. Code: https://github.com/ soroushseifi/glimpse-attend-explore.

show abstract

“…Some of the earliest work [1] provides a general framework for this problem in low resource vision systems and camera control [5]. Recent work in this domain aims at learning view selection strategies to solve diverse tasks including object recognition [29,2], segmentation [8,28], visual navigation [39,10,44], and pose estimation [16,37]. Similar work to ours falls in the domain of active image understanding, in the subsequent subsection we provide a brief literature survey of the constituent modules of our model.…”

Section: Related Workmentioning

confidence: 99%

Glimpse-Attend-and-Explore: Self-Attention for Active Visual Exploration

Seifi

¹

,

Jha

²

,

Tuytelaars

³

2021

2021 IEEE/CVF International Conference on Computer Vision (ICCV)

View full text Add to dashboard Cite

Active visual exploration aims to assist an agent with a limited field of view to understand its environment based on partial observations made by choosing the best viewing directions in the scene. Recent methods have tried to address this problem either by using reinforcement learning, which is difficult to train, or by uncertainty maps, which are task-specific and can only be implemented for dense prediction tasks. In this paper, we propose the Glimpse-Attendand-Explore model which: (a) employs self-attention to guide the visual exploration instead of task-specific uncertainty maps; (b) can be used for both dense and sparse prediction tasks; and (c) uses a contrastive stream to further improve the representations learned. Unlike previous works, we show the application of our model on multiple tasks like reconstruction, segmentation and classification. Our model provides encouraging results while being less dependent on dataset bias in driving the exploration. We further perform an ablation study to investigate the features and attention learned by our model. Finally, we show that our self-attention module learns to attend different regions of the scene by minimizing the loss on the downstream task. Code: https://github.com/ soroushseifi/glimpse-attend-explore.

show abstract

Deep Reinforcement Learning for Active Human Pose Estimation

Cited by 27 publications

References 18 publications

Applied Artificial Intelligence and user satisfaction: Smartwatch usage for healthcare in Bangladesh during COVID-19

Applied Artificial Intelligence and user satisfaction: Smartwatch usage for healthcare in Bangladesh during COVID-19

Glimpse-Attend-and-Explore: Self-Attention for Active Visual Exploration

Glimpse-Attend-and-Explore: Self-Attention for Active Visual Exploration

Contact Info

Product

Resources

About