Post-processing of wall and floor pose recognition and identification
The objective of this research activity is the development of a post-processing algorithm for the improvement of segmentation and 3D floor and wall pose recognition from RGB images. We will develop an automatic post-processing model on the dataset provided by the customer. We will use datasets available in the literature for both the development and the experimental testing of the prototype, with particular reference to the environments present on the Habitat platform (https://aihabitat.org/).
With 30 years of experience, Maticad S.r.l. is an Italian excellence in the field of software dedicated to Interior Design. The company is able to propose innovative, complete and versatile solutions for manufacturers, distributors, designers and customers, anywhere in the world. Maticad operates in the field of software development for systems and devices in the field of ceramic tiles and wall covering materials, sanitary ware, bathroom furniture and interior design.
The company needs to develop a system for post-processing the results of a floor and wall segmentation algorithm from RGB images and it decided to hire the Artificial Intelligence Research and Innovation Center (AIRI) for the task.
From this idea comes the collaboration between AImageLab and Maticad, resulting in the research programme entitled “Post-processing of wall and floor pose recognition and identification”. The research programme consists in the realisation of a technique for the post-processing and refinement of the predictions of an algorithm for the segmentation and recognition of the 3D laying of walls and floors from RGB images, in the testing and experimental validation on datasets provided by the client and on existing datasets in the literature. The company provides us with a dataset consisting of about 2000 manually annotated images with information about the 3D laying of floors and/or walls. We will provide the company with a package containing the automatic model created, the relevant weights or parameters, and a demonstrator code showing how the model works on example images.