DigiArt consortium members CNRS, are developing a content based 3D shape retrieval system for non-expert users and low cost sensors. The National Center for Scientific Research is one of the largest research institutions in the world. Internationally recognized for the excellence of its scientific work, the CNRS is a reference in the world of research and development as well as for the general public.
They say, “Our CB3DR solution allows to scan an object on the fly (with a free manual acquisition) and retrieve similar shapes from a database.”
“First the user scans an object with a 3D low-cost scanner, finding the most similar shapes by a similarity search. Then, we present an interactive learning algorithm based on Support Vector Machine which requires very simple positive-negative annotations from the users. We have evaluated our system on several standard 3D datasets (Cat10, Cat31, Cat60, RGB-D, ModelNet10, and our own low-resolution shape dataset made with the manual Occipital Structure Sensor considering multiple state-of-the-art descriptors. We investigate how point cloud resolution can affect the results and show that our framework can deal with different quality acquisition sensors.”
“The final goal of this work is to provide a framework for 3D object acquisition and retrieval which should be manageable by non-experts such as, for instance, archaeologists or paleontologists when they are on an archaeological site and they would like to quickly and easily scan an artifact in 3D to query it in remote large 3D object databases.”
“So far, we have brought three main contributions to the community:
First, we made an extensive evaluation of global descriptors for representing 3D shape point cloud object. Then, we built a 3D search engine based on similarity search and interactive learning, and finally, we evaluate how our system deals with a 3D point cloud acquired by a low-cost device as a shape query and retrieve its category on a 3D CAD object dataset and conversely.”
Below, you will find a video showing how our interactive Learning scheme works:
And below you will find a video showing how we can retrieve objects directly from the iPad by scanning partial or full views