GQKD: Greedy Query Knowledge Distillation for Efficient 3D Instance Segmentation
IEEE Signal Processing Letters · 2026 · 33 · 1801-1805
DOI: 10.1109/lsp.2026.3685568 ↗Kurz gefasst
Fachartikel in der Zeitschrift IEEE Signal Processing Letters, 2026. Verfasst von Taehyun Jeong und 3 weiteren Autor:innen. Die Arbeit nennt 28 Literaturangaben. Gefördert von: Institute of Information & Communications Technology Planning & Evaluation, Korea Government, Development of 3D Space Digital Media Standard Technology.
Abstract (Originalsprache)
3D instance segmentation is a fundamental component of various 3D scene understanding applications. Recent transformer-based approaches have achieved strong performance through query-driven decoding. However, their repeated multi-layer decoder architectures impose substantial computational overhead, motivating the need for model compression. In this work, we identify two key phenomena across decoder layers: object disappearance, where valid object queries vanish as decoding proceeds, and object degradation, where intermediate-layer predictions can be more reliable than final-layer outputs. Based on these observations, we propose Greedy Query Knowledge Distillation (GQKD), an effective distillation framework for lightweight 3D instance segmentation. GQKD constructs a pseudo-GT query set to supervise the student model by selecting the best true-positive query for each instance across multiple decoder layers, along with low-confidence negative teacher queries. This Pseudo-GT recovers high-quality object hypotheses lost during decoding and provides balanced supervision by supplementing recovered positives with low-confidence negatives. Experimental results on ScanNetV2 demonstrate that GQKD enables a student model with approximately 50% reduced decoder depth to achieve performance close to the teacher when used alone, and further improves results when combined with existing distillation strategies.
Förderung
- Institute of Information & Communications Technology Planning & Evaluation
- Korea Government
- Development of 3D Space Digital Media Standard Technology
- Industrial Core Technology Development Program of MOTIE/KEIT
Zitieren
APA
Jeong, T., Moon, H., Lee, S., & Jeong, J. (2026). GQKD: Greedy Query Knowledge Distillation for Efficient 3D Instance Segmentation. IEEE Signal Processing Letters, 33, 1801-1805. https://doi.org/10.1109/lsp.2026.3685568
Vancouver
Jeong T, Moon H, Lee S, Jeong J. GQKD: Greedy Query Knowledge Distillation for Efficient 3D Instance Segmentation. IEEE Signal Processing Letters. 2026;33:1801-1805. doi:10.1109/lsp.2026.3685568
BibTeX
@article{Jeong2026,
author = {Jeong, Taehyun and Moon, Hyeon-Cheol and Lee, Suhyeon and Jeong, Jinwoo},
title = {GQKD: Greedy Query Knowledge Distillation for Efficient 3D Instance Segmentation},
journal = {IEEE Signal Processing Letters},
year = {2026},
volume = {33},
pages = {1801--1805},
doi = {10.1109/lsp.2026.3685568}
}
EUVIMED