| 1. |
- Brissman, Emil, et al.
(författare)
-
Predicting Signed Distance Functions for Visual Instance Segmentation
- 2021
-
Ingår i: 33rd Annual Workshop of the Swedish-Artificial-Intelligence-Society (SAIS). - : Institute of Electrical and Electronics Engineers (IEEE). - 9781665442367 - 9781665442374 ; , s. 5-10
-
Konferensbidrag (refereegranskat)abstract
- Visual instance segmentation is a challenging problem and becomes even more difficult if objects of interest varies unconstrained in shape. Some objects are well described by a rectangle, however, this is hardly always the case. Consider for instance long, slender objects such as ropes. Anchor-based approaches classify predefined bounding boxes as either negative or positive and thus provide a limited set of shapes that can be handled. Defining anchor-boxes that fit well to all possible shapes leads to an infeasible number of prior boxes. We explore a different approach and propose to train a neural network to compute distance maps along different directions. The network is trained at each pixel to predict the distance to the closest object contour in a given direction. By pooling the distance maps we obtain an approximation to the signed distance function (SDF). The SDF may then be thresholded in order to obtain a foreground-background segmentation. We compare this segmentation to foreground segmentations obtained from the state-of-the-art instance segmentation method YOLACT. On the COCO dataset, our segmentation yields a higher performance in terms of foreground intersection over union (IoU). However, while the distance maps contain information on the individual instances, it is not straightforward to map them to the full instance segmentation. We still believe that this idea is a promising research direction for instance segmentation, as it better captures the different shapes found in the real world.
|
|
| 2. |
- Cooney, Martin, 1980-, et al.
(författare)
-
Robot First Aid : Autonomous Vehicles Could Help in Emergencies
- 2021
-
Ingår i: 2021 Swedish Artificial Intelligence Society Workshop (SAIS). - : IEEE. - 9781665442367 - 9781665442374
-
Konferensbidrag (refereegranskat)abstract
- Safety is of critical importance in designing autonomous vehicles (AVs) that will be able to perform effectively in complex, mixed-traffic, real-world urban environments. Some prior research has looked at how to proactively avoid accidents with safe distancing and driver monitoring, but currently little research has explored strategies to recover afterwards from emergencies, from crime to natural disasters. The current short paper reports on our ongoing work using a speculative prototyping approach to explore this expansive design space, in the context of how a robot inside an AV could be deployed to support first aid. As a result, we present some proposals for how to detect emergencies, and examine and help victims, as well as lessons learned in prototyping. Thereby, our aim is to stimulate discussion and ideation that-by considering the prevalence of Murphy's law in our complex world, and the various technical, ethical, and practical concerns raised-could potentially lead to useful safety innovations. © 2021 IEEE.
|
|
| 3. |
- Guerrero, Esteban, et al.
(författare)
-
Towards motivation-driven intelligent interfaces: Formal argumentation meets activity theory
- 2021
-
Ingår i: 2021 Swedish Artificial Intelligence Society Workshop (SAIS). - : IEEE. - 9781665442367
-
Konferensbidrag (refereegranskat)abstract
- Theories about human activity and motivation point out that motives are driving forces behind human activities and development of healthy and unhealthy habits. Activity theory is one of these that has been applied to develop activity-centered user interfaces. Activity theory differentiates between sense-making and stimuli-oriented types of motives that have a strong influence on our daily behavior. Two main challenges are explored in this paper: 1) the personalisation of graphical user interfaces to mediate representations of motivation-based activities to support behaviour change processes; and 2) the proactiveness of such visual representations.As methods, we use activity theory as a framework for defining the motivations' dynamics, and formal argumentation theory as the underlying mechanism for interactive reasoning and decision-making in the process of generating the user interface.Our contributions are two-folded: 1) a dynamic graphical user interface where the background responds to behaviors linked to sense-making motives, and the foreground to stimuli motivation; and 2) a non-monotonic reasoning mechanism endowing the user interface with proactiveness (not only react to the user interactions but trigger and direct attention to potential conflicts), and a motive-based behavior conflict resolution process. Future work includes user studies to explore how triggering of focus may create increased awareness in an individual of conflicting motives in daily activities and how this may support changes of unhealthy habits.
|
|
| 4. |
- Holmquist, Karl, 1992-, et al.
(författare)
-
Class-Incremental Learning for Semantic Segmentation - A study
- 2021
-
Ingår i: 2021 Swedish Artificial Intelligence Society Workshop (SAIS). - : IEEE. - 9781665442367 - 9781665442374 ; , s. 25-28
-
Konferensbidrag (refereegranskat)abstract
- One of the main challenges of applying deep learning for robotics is the difficulty of efficiently adapting to new tasks while still maintaining the same performance on previous tasks. The problem of incrementally learning new tasks commonly struggles with catastrophic forgetting in which the previous knowledge is lost.Class-incremental learning for semantic segmentation, addresses this problem in which we want to learn new semantic classes without having access to labeled data for previously learned classes. This is a problem in industry, where few pre-trained models and open datasets matches exactly the requisites. In these cases it is both expensive and labour intensive to collect an entirely new fully-labeled dataset. Instead, collecting a smaller dataset and only labeling the new classes is much more efficient in terms of data collection.In this paper we present the class-incremental learning problem for semantic segmentation, we discuss related work in terms of the more thoroughly studied classification task and experimentally validate the current state-of-the-art for semantic segmentation. This lays the foundation as we discuss some of the problems that still needs to be investigated and improved upon in order to reach a new state-of-the-art for class-incremental semantic segmentation.
|
|
| 5. |
- Kock, Elina, et al.
(författare)
-
Identifying cheating behaviour with machine learning
- 2021
-
Ingår i: 33rd Workshop of the Swedish Artificial Intelligence Society, SAIS 2021. - : Institute of Electrical and Electronics Engineers Inc.. - 9781665442367
-
Konferensbidrag (refereegranskat)abstract
- We have investigated machine learning based cheating behaviour detection in physical activity-based smart-phone games. Sensor data were acquired from the accelerometer/gyroscope of an iPhone 7 during activities such as jumping, squatting, stomping, and their cheating counterparts. Selected attributes providing the most information gain were used together with a sequential model yielding promising results in detecting fake activities. Even better results were achieved by employing a random forest classifier. The results suggest that machine learning is a strong candidate for detecting cheating behaviours in physical activity-based smartphone games.
|
|
| 6. |
- Peretz-Andersson, Einav, et al.
(författare)
-
AI Transformation in the Public Sector : Ongoing Research
- 2021
-
Ingår i: 33rd Workshop of the Swedish Artificial Intelligence Society, SAIS 2021. - : Institute of Electrical and Electronics Engineers (IEEE). - 9781665442367 ; , s. 33-36
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- Real-world application of data-driven and intelligent systems (AI) is increasing in the private and public sector as well as in society at large. Many organizations transform as a consequence of increased AI implementation. The consequences of such transformations may include new recruitment plans, procurement of additional IT, changes in existing positions and roles, new business models, as well as new policies and regulations. However, it is unclear how this transformation varies across different types of organizations. We study the effects of bottom-up approaches, such as pilot projects and mentoring to specific groups within organizations, and aim to explore how such approaches can complement the top-down approach of strategic AI implementation. Our context is the public sector. Our goal is to acquire an improved understanding of how and when AI transformation occurs in the public sector, which are the consequences, and which strategies are fruitful or detrimental to the organization. We aim to study public sector organizations in Sweden, Norway, New Zealand, Germany, and The Netherlands to learn about potential similarities and differences with regard to AI transformation.
|
|
| 7. |
- Stenhager, E., et al.
(författare)
-
Hit Detection in Sports Pistol Shooting
- 2021
-
Ingår i: 33rd Workshop of the Swedish Artificial Intelligence Society, SAIS 2021. - : Institute of Electrical and Electronics Engineers (IEEE). - 9781665442367 ; , s. 42-45
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- Score calculation and performance analysis of shooting targets is an important aspect in the development of sports shooting ability. An image-based automatic scoring algorithm would provide automation of this procedure and digital visualization of the result. Existing solutions are able to detect hits with high precision. However, these methods are either too expensive or adapted to unrealistic use cases where high quality paper targets are photographed in very favorable environments. Usually, precision pistol shooting is performed outdoors and bullet holes are covered with stickers between shooting rounds. The targets are reused until they are destroyed. This paper introduces the first generation of an image-based method for automatic hit detection adapted to realistic shooting conditions. It relies solely on available image processing techniques. The proposed algorithm detects hits with 40 percent detection rate in low-quality targets, reaching 88 percent detection rate in targets of higher quality.
|
|
| 8. |
- Wang, Qinghua, et al.
(författare)
-
Smart Sewage Water Management and Data Forecast
- 2021
-
Ingår i: 33rd Workshop of the Swedish Artificial Intelligence Society, SAIS 2021. - USA : Institute of Electrical and Electronics Engineers (IEEE). - 9781665442367
-
Konferensbidrag (övrigt vetenskapligt/konstnärligt)abstract
- There is currently an ongoing digital transformation for sewage and wastewater management. By automating data collection and enabling remote monitoring, we will not only be able to save abundant human resources but also enabling predictive maintenance which is based on big data analytics. This paper presents a smart sewage water management system which is currently under development in southern Sweden. Real-time data can be collected from over 500 sensors which have already been partially deployed. Preliminary data analysis shows that we can build statistical data models for ground water, rainfall, and sewage water flows, and use those models for data forecast and anomaly detection.
|
|
