GeoMask3D: Geometrically Informed Mask Selection for Self-Supervised Point Cloud Learning in 3D

Ali Bahri, Moslem Yazdanpanah, Mehrdad Noori, Milad Cheraghalikhani, Gustavo Adolfo Vargas Hakim, David OSOWIECHI, Farzad Beizaee, Ismail Ben Ayed, Christian Desrosiers

Abstract:

We introduce a novel approach to self-supervised learning for point clouds, employing a geometrically informed mask selection strategy called GeoMask3D (GM3D) to boost the efficiency of Masked Auto Encoders (MAE). Unlike the conventional method of random
masking, our technique utilizes a teacher-student model to focus on intricate areas within the data, guiding the model’s focus toward regions with higher geometric complexity. This strategy is grounded in the hypothesis that concentrating on harder patches yields a more
robust feature representation, as evidenced by the improved performance on downstream tasks. Our method also presents a feature-level knowledge distillation technique designed to guide the prediction of geometric complexity, which utilizes a comprehensive context from
feature-level information. Extensive experiments confirm our method’s superiority over State-Of-The-Art (SOTA) baselines, demonstrating marked improvements in classification, segmentation, and few-shot tasks.

Submission Length: Regular submission (no more than 12 pages of main content)

Changes Since Last Submission:

In the final version:

1. We have revised the writing to enhance the presentation of the proposed geometrically informed mask selection strategy and its relationship to Geometric Complexity. These improvements have been incorporated into Sections 3.2 and 3.2.1 of the main paper.

2. Based on the rebuttal phase, we have added an experiment on the integration of GeoMask3D into Point-FEMAE in the supplementary material (Section 3).

3. From the rebuttal phase, we have included an experiment analyzing the impact of geometrically guided masking compared to random masking in the supplementary material (Section 3).

4. Based on the rebuttal phase, we have added a time analysis section in the supplementary material (Section 4).

5. The results in Table 5 of the original paper have been updated based on our response to Question 2 from Reviewer JYxP in the rebuttal.

Code: https://github.com/AliBahri94/GM3D

Assigned Action Editor: ~Wei_Liu3

Submission Number: 3234