Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Intent: INteraction TENdency towards Targets
From robotics to human-computer interaction, there are numerous real-world tasks that would benefit from practical systems that can anticipate future high-level actions and predict intention and goals based on observation of the past. Intention prediction is not only important for care robots to anticipate people’s actions but is also a key challenge in the design of artificial intelligent systems.
Keywords: intent prediction, interaction, eye tracking, vision language
The main objective of this project is to learn how people predict others’ intent by observing where they observe. Conventional intention prediction systems rely on dense annotations (action labels, segmentation masks etc.) collected from humans, yet they still struggle with accurately predicting actions in the next few seconds. In fact, both recognition of complex actions as well as efficient learning from video data are active research topics The main objective of this project is to learn how people predict others’ intent by observing where they observe. Conventional intention prediction systems rely on dense annotations (action labels, segmentation masks etc.) collected from humans, yet they still struggle with accurately predicting actions in the next few seconds. In fact, both recognition of complex actions as well as efficient learning from video data are active research topics and remain challenging. In this project, we aim to learn how people understand the underlying goal of the task by inferring their strategy from the information they seek and from their prediction of future interactions. Don’t hesitate to reach out to us if you find the general direction interesting or have any questions.
Benefits:
- Gain research experience in hot topics in both computer vision with the potential to interact with leading experts in the fields;
- Hands-on experience in applying state-of-the-art machine learning techniques in solving real-world problems;
- Possibility to publish in top-tier venues as the lead author;
- Fulfill course requirements.
Requirements:
Interested students should have a background in computer vision and deep learning, and have a working knowledge of available learning frameworks. This project is highly research-oriented, and we encourage motivated students who are interested in gaining more research experience to directly contact us via email. Excellent outcomes can easily result in a publication in top-tire venues.
Literature:
Watch-And-Help: A Challenge for Social Perception and Human-AI Collaboration
https://arxiv.org/abs/2010.09890
Actor and Observer: Joint Modeling of First and Third-Person Videos
https://arxiv.org/pdf/1804.09627.pdf
The main objective of this project is to learn how people predict others’ intent by observing where they observe. Conventional intention prediction systems rely on dense annotations (action labels, segmentation masks etc.) collected from humans, yet they still struggle with accurately predicting actions in the next few seconds. In fact, both recognition of complex actions as well as efficient learning from video data are active research topics The main objective of this project is to learn how people predict others’ intent by observing where they observe. Conventional intention prediction systems rely on dense annotations (action labels, segmentation masks etc.) collected from humans, yet they still struggle with accurately predicting actions in the next few seconds. In fact, both recognition of complex actions as well as efficient learning from video data are active research topics and remain challenging. In this project, we aim to learn how people understand the underlying goal of the task by inferring their strategy from the information they seek and from their prediction of future interactions. Don’t hesitate to reach out to us if you find the general direction interesting or have any questions.
Benefits:
- Gain research experience in hot topics in both computer vision with the potential to interact with leading experts in the fields;
- Hands-on experience in applying state-of-the-art machine learning techniques in solving real-world problems;
- Possibility to publish in top-tier venues as the lead author;
- Fulfill course requirements.
Requirements:
Interested students should have a background in computer vision and deep learning, and have a working knowledge of available learning frameworks. This project is highly research-oriented, and we encourage motivated students who are interested in gaining more research experience to directly contact us via email. Excellent outcomes can easily result in a publication in top-tire venues.
Literature:
Watch-And-Help: A Challenge for Social Perception and Human-AI Collaboration https://arxiv.org/abs/2010.09890
Actor and Observer: Joint Modeling of First and Third-Person Videos https://arxiv.org/pdf/1804.09627.pdf
Not specified.
Not specified.
Xi Wang (xi.wang@inf.ethz.ch)
Emre Aksan (eaksan@inf.ethz.ch)