MVTN: Multi-View Transformation Network for 3D Shape Recognition

By Abdullah Hamdi, Silvio Giancola, Bing Li, Ali K. Thabet, Bernard Ghanem. IEEE International Conference on Computer Vision, 2020

298 citations recorded by the source corpus.

Multi-view projection methods have demonstrated their ability to reach state-of-the-art performance on 3D shape recognition. Those methods learn different ways to aggregate information from multiple views. However, the camera view-points for those views tend to be heuristically set and fixed for all shapes. To circumvent the lack of dynamism of current multi-view methods, we propose to learn those view-points. In particular, we introduce the Multi-View Transformation Network (MVTN) that regresses optimal view-points for 3D shape recognition, building upon advances in differentiable rendering. As a result, MVTN can be trained end-to-end along with any multi-view network for 3D shape classification. We integrate MVTN in a novel adaptive multi-view pipeline that can render either 3D meshes or point clouds. MVTN exhibits clear performance gains in the tasks of 3D shape classification and 3D shape retrieval with-out the need for extra training supervision. In these tasks, MVTN achieves state-of-the-art performance on ModelNet40, ShapeNet Core55, and the most recent and realistic ScanObjectNN dataset (up to 6% improvement). Interestingly, we also show that MVTN can provide network robustness against rotation and occlusion in the 3D domain. The code is available at https://github.com/ajhamdi/MVTN.

Subfields: Computational Mechanics, Computer Vision and Pattern Recognition, Geology

View Paper: https://www.semanticscholar.org/paper/7e6c7cc0ac0b0d03fa4790f3927f8bc9fee4fb66

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Computational Mechanics

1. Guocheng Qian
2. Yu Qiao
3. Bin Cui
4. Peng Gao
5. Dieter Fox
6. Hasan Hammoud
7. Jinjie Mai
8. Mohamed Elhoseiny

In Computer Vision and Pattern Recognition

1. Yu Qiao
2. Hongsheng Li
3. Luc Van Gool
4. Xiaogang Wang
5. Dacheng Tao
6. Jitendra Malik
7. Trevor Darrell
8. Ming-Hsuan Yang

In Geology

1. Guocheng Qian
2. Yu Qiao
3. Peng Gao
4. Dieter Fox
5. Hasan Hammoud
6. Jinjie Mai
7. Mohamed Elhoseiny
8. Renrui Zhang

Citations in the corpus

• PointCLIP: Point Cloud Understanding by CLIP (2021, methodology): “With 3D sensing technology developing rapidly, the growing demand for processing 3D point cloud data has boosted many advanced deep models with better local feature aggregator [30, 32, 50], geometry modeling [20, 40, 60] and projection-based processing [21, 34, 49].”
• Rethinking Network Design and Local Geometry in Point Cloud: A Simple Residual MLP Framework (2022, unknown): “Even compared with the heavy multi-view projection method MVTN (12 views), our PointMLP still performs much better (85.39% 82.8% ).”
• PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (2022, methodology): “However, recent works [12, 28, 34] show an increasing interest in the real-world scanned dataset ScanObejectNN compared to this synthesized dataset.”
• RVT: Robotic View Transformer for 3D Object Manipulation (2023, methodology): “Follow-up works like MVTN [53] and Voint cloud [54] further improved upon SimpleView.”