ICM-3D: Instantiated Category Modeling for 3D Instance Segmentation

TL;DR

ICM-3D introduces a unified, single-step approach for 3D instance segmentation by reformulating it as a per-point classification problem, leveraging instantiated categorization from spatial data.

Contribution

The paper presents ICM-3D, a novel integrated method that simplifies 3D instance segmentation into a single-step process, improving over traditional multi-step pipelines.

Findings

Achieves high performance across multiple benchmarks.

Effective across various frameworks and backbones.

Outperforms existing two-step methods.

Abstract

Separating 3D point clouds into individual instances is an important task for 3D vision. It is challenging due to the unknown and varying number of instances in a scene. Existing deep learning based works focus on a two-step pipeline: first learn a feature embedding and then cluster the points. Such a two-step pipeline leads to disconnected intermediate objectives. In this paper, we propose an integrated reformulation of 3D instance segmentation as a per-point classification problem. We propose ICM-3D, a single-step method to segment 3D instances via instantiated categorization. The augmented category information is automatically constructed from 3D spatial positions. We conduct extensive experiments to verify the effectiveness of ICM-3D and show that it obtains inspiring performance across multiple frameworks, backbones and benchmarks.