Summary of Inemo: Incremental Neural Mesh Models For Robust Class-incremental Learning, by Tom Fischer et al.
iNeMo: Incremental Neural Mesh Models for Robust Class-Incremental Learning
by Tom Fischer, Yaoyao Liu, Artur Jesslen, Noor Ahmed, Prakhar Kaushik, Angtian Wang, Alan Yuille, Adam Kortylewski, Eddy Ilg
First submitted to arxiv on: 12 Jul 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A novel approach to continual learning is proposed, leveraging the strong performance of non-continual neural mesh models in generalizing out-of-distribution (OOD) scenarios. The incremental neural mesh model can be extended with new meshes over time, and a latent space initialization strategy allocates feature space for future unseen classes in advance. A positional regularization term ensures features stay within respective regions. Extensive experiments on Pascal3D and ObjectNet3D datasets show the approach outperforms baselines by 2-6% in-domain and 6-50% OOD, with a first incremental learning approach for pose estimation. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary A new way to learn continuously is developed, which takes advantage of neural mesh models’ ability to handle unknown data. This method can be updated over time and prepares for new classes by setting up the right space beforehand. It also keeps features in the correct places. Tests on Pascal3D and ObjectNet3D show this approach works better than others by 2-6% within the same dataset and 6-50% when dealing with unknown data, and it’s the first to work for pose estimation. |
Keywords
» Artificial intelligence » Continual learning » Latent space » Pose estimation » Regularization
