RETech'18- Proceedings of the 2018 ACM Workshop on Multimedia for Real Estate Tech

Apartment searching has been a hot demand all over the world. Understanding the apartment structure will significantly contribute to simplifying the searching process. In this project, the fully convolutional networks are applied to generate semantic segmentation for the apartment floor plan images. We then study and optimize an algorithm to extract the graph model from the segmentation, and to extract the maximum common subgraph with which the structure similarity can be measured. We generate the ground truth data by using online annotation tool and experiment on half of the data. Our segmentation prediction results achieve a pixel mean accuracy of $0.89$ for the full-label model, and retrieve quite similar apartment structures by using graph modeling.

In recent years, with the progress of e-commerce, recommendation for not only mass-produced daily items, such as books, but also special items that are not mass-produced has become an important task. In this study, we present an algorithm for real estate recommendation. There are no identical properties in the world, properties already occupied by someone else cannot be recommended, and users rent or buy properties only a few times in their lives. Therefore, automatic property recommendation is one of the most difficult tasks. In this study, we predict users' preference for properties, which is the first step of property recommendation, by combining content-based filtering and multilayer perceptron (MLP). In the MLP, we used not only attribute data of users and properties but also the deep features extracted from floor plan images of properties. As a result, we succeeded in predicting users' preference with an accuracy of 60.7%.

The condition of a building is an important factor for real estate valuation. Currently, the estimation of condition is determined by real estate appraisers which makes it subjective to a certain degree. We propose a novel vision-based approach for the assessment of the building condition from exterior views of the building. To this end, we develop a multi-scale patch-based pattern extraction approach and combine it with convolutional neural networks to estimate building condition from visual clues. Our evaluation shows that visually estimated building condition can serve as a proxy for condition estimates by appraisers.

We consider the problem of depth reconstruction from downsampled sparse depth values. We compare our approach with semi-dense depth map interpolation and direct RGB-to-Depth reconstruction solutions on several datasets, including Matterport 3D dataset containing RGB and depth images of 90 building-scale scenes. We demonstrate that the proposed model can produce approximate depth map for over two hundreds images per second.

Due to the rapid development of the Internet, it became easy to hold a teleconference. However, even in current teleconference systems, it is difficult for attendees of teleconference to share a feeling that they are at common space. On the other hand, even though a mode of transportation is quite developed, it is still hard to move so long distance for attending a conference every time. These situations indicate one needs, that is, new teleconference system by which the attendee can feel that they are existing at the same room. So, in this paper, we will propose a new framework to make a teleconference system for sharing living-in-room by using VR technology. In our system, we use Mixed Reality (MR) and Virtual Reality (VR) technologies. A user using an MR device recognizes another user at a remote place as an avatar, and a user using a VR device experiences a video of a remote place as a virtual space. As described above, each user experiences a sense as if they are in the same space. In addition, this feeling is called "feeling of sharing living-in-room" in our study. We conducted experiments on the proposed system. We divided the VR-side user and the MR-side user into the separate condition and evaluated it in terms of the feeling of sharing living-in-room users got. As a result, we showed that the VR-side user was effective in getting the feeling of sharing living-in-room by moving the viewpoint. The presence of avatars is currently implemented in the system was not effective for fostering the feeling of sharing living-in-room, and from this result, we found several points of the system for improvement related to avatars.