![[2412.02039] Multi-View 3D Reconstruction using Knowledge Distillation](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2412.02039] Multi-View 3D Reconstruction using Knowledge Distillation
Summary
This paper presents a knowledge distillation approach for Multi-View 3D reconstruction, utilizing a teacher-student model framework to enhance efficiency and performance in scene representation.
Why It Matters
As 3D reconstruction technologies advance, optimizing their performance while reducing computational demands is crucial. This research addresses the challenges of inference time and resource usage, making it relevant for applications in computer vision and AI.
Key Takeaways
- The study introduces a knowledge distillation pipeline to improve 3D reconstruction efficiency.
- Two architectures, CNN and Vision Transformer, were tested, with the latter showing superior performance.
- The research highlights the importance of scene-specific representations in model training.
Computer Science > Computer Vision and Pattern Recognition arXiv:2412.02039 (cs) [Submitted on 2 Dec 2024 (v1), last revised 18 Feb 2026 (this version, v2)] Title:Multi-View 3D Reconstruction using Knowledge Distillation Authors:Aditya Dutt, Ishikaa Lunawat, Manpreet Kaur View a PDF of the paper titled Multi-View 3D Reconstruction using Knowledge Distillation, by Aditya Dutt and 1 other authors View PDF HTML (experimental) Abstract:Large Foundation Models like Dust3r can produce high quality outputs such as pointmaps, camera intrinsics, and depth estimation, given stereo-image pairs as input. However, the application of these outputs on tasks like Visual Localization requires a large amount of inference time and compute resources. To address these limitations, in this paper, we propose the use of a knowledge distillation pipeline, where we aim to build a student-teacher model with Dust3r as the teacher and explore multiple architectures of student models that are trained using the 3D reconstructed points output by Dust3r. Our goal is to build student models that can learn scene-specific representations and output 3D points with replicable performance such as Dust3r. The data set we used to train our models is 12Scenes. We test two main architectures of models: a CNN-based architecture and a Vision Transformer based architecture. For each architecture, we also compare the use of pre-trained models against models built from scratch. We qualitatively compare the reconstructed...