Abstract | ||
---|---|---|
We present FlowNet3D++, a deep scene flow estimation network. Inspired by classical methods, FlowNet3D++ incorporates geometric constraints in the form of point-toplane distance and angular alignment between individual vectors in the flow field, into FlowNet3D [21]. We demonstrate that the addition of these geometric loss terms improves the previous state-of-art FlowNet3D accuracy from 57.85% to 63.43%. To further demonstrate the effectiveness of our geometric constraints, we propose a benchmark for flow estimation on the task of dynamic 3D reconstruction, thus providing a more holistic and practical measure of performance than the breakdown of individual metrics previously used to evaluate scene flow. This is made possible through the contribution of a novel pipeline to integrate point-based scene flow predictions into a global dense volume. FlowNet3D++ achieves up to a 15.0% reduction in reconstruction error over FlowNet3D, and up to a 35.2% improvement over KillingFusion [32] alone. We will release our scene flow estimation code later. |
Year | DOI | Venue |
---|---|---|
2020 | 10.1109/WACV45572.2020.9093302 | 2020 IEEE Winter Conference on Applications of Computer Vision (WACV) |
Keywords | DocType | ISSN |
geometric constraints,geometric loss,dynamic 3D reconstruction,point-based scene flow predictions,FlowNet3D++,deep scene flow estimation code | Conference | 2472-6737 |
ISBN | Citations | PageRank |
978-1-7281-6554-7 | 5 | 0.39 |
References | Authors | |
22 | 5 |
Name | Order | Citations | PageRank |
---|---|---|---|
Wang Zirui | 1 | 5 | 0.39 |
Li Shuda | 2 | 5 | 0.72 |
Howard-Jenkins Henry | 3 | 5 | 1.06 |
Victor Adrian Prisacariu | 4 | 465 | 25.75 |
Min Chen | 5 | 1293 | 82.69 |