A real-time, robust and flexible visual recognition system capable to cope with shape- and materials-diversity requires:
- Hybrid representations
- Multimodal recognition methods
- Large and high quality training and validation data sets
The solution to these problems have off-line and on-line phases. The focus of this presentation is placed on various elements of the off-line phase. The talk introduces a minimal-cost fully automatic method to generate 3D object models with unlimited resolution from a single aligned image for arbitrary revolution objects. The scope is focused on objects with complex (non-lambertian) materials that have intricate free-form revolution surfaces. The approach is solely based on a single 2D image extracting two cues: The object's external contour and internal skeleton. This method achieves unprecedented high quality 3D models with ultra-efficient serialization. Claims are supported by extensive evaluation using data sets with ground-truth metrics for volume, area, file-size, and execution time.
Presented at the 11th Bay Area Multimedia Forum (BAMMF), 13 December 2016 (bammf.org)