Research map / Continual learning / Class-incremental
iNeMo: Incremental Neural Mesh Models for Robust Class-Incremental Learning
Tom Fischer, Yaoyao Liu, Artur Jesslen, Noor Ahmed, Prakhar Kaushik, Angtian Wang, Alan Yuille, Adam Kortylewski, Eddy Ilg
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add pose-aware exemplar selection instead of herding (ablation) | iNeMo · P3D B0+3 | final accuracy: 93.60 -> 94.70 | Improved |
| add latent space initialization (ablation) | iNeMo · O3D B0+20 | final accuracy: 70.32 -> 71.67 | Improved |
| add positional regularization (ablation) | iNeMo · P3D B0+3 | final accuracy: 94.78 -> 94.98 | Improved |
| add knowledge distillation to the full latent-space setup (ablation) | iNeMo · O3D B0+20 | final accuracy: 72.09 -> 80.17 | Improved |
| add uncertainty term to classification inference (ablation) | iNeMo · O3D B0+20 | final accuracy: 75.8 -> 80.17 | Improved |
| fine-tune NeMo incrementally without replay or distillation (ablation) | NeMo · P3D B0+3 | final accuracy: 17.47 | Hurt |
| apply LwF to NeMo instead of ResNet50 (ablation) | LwF · P3D B0+3 | final accuracy: 83.75 -> 17.45 | Hurt |
| add exemplar replay (iCaRL herding) to NeMo (ablation) | NeMo · P3D B0+3 | final accuracy: 17.45 -> 93.72 | Improved |
| reduce exemplars per class from 20 to 5 (ablation) | iNeMo · P3D B0+3 | final accuracy: 96.41 -> 82.59 | Hurt |
| add pose supervision head to ResNet50 baseline (ablation) | iCaRL ResNet50 · P3D B0+3 | final accuracy: 91.79 -> 89.86 | Hurt |
| use NeMo representation instead of ResNet50 for incremental classification | iNeMo vs 2D baselines · Occluded-P3D L3 | accuracy: 55.89 -> 71.55 | Improved |
| iNeMo vs 2D baselines for pose pi/6 under strong corruption | iNeMo · Corrupted-P3D pose pi/6 | accuracy: 47.30 vs 54.57 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- iCaRL — iCaRL: Incremental Classifier and Representation Learning (CVPR 2017) · better on 3 of 3 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Class-Incremental Learning: Survey and Performance Evaluation on Image Classification (TPAMI 2022 · similarity 0.86)
- Geometry-driven OOD Detectors Are Class-Incremental Learners (CVPR 2026 · similarity 0.85)
- Point-UQ: An Uncertainty-Quantification Paradigm for Point Cloud Few-Shot Class Incremental Learning (ICLR 2026 · similarity 0.85)
- Advancing Cross-domain Discriminability in Continual Learning of Vision-Language Models (NeurIPS 2024 · similarity 0.84)
- Overcoming the Pitfalls of Vision-Language Model Finetuning for OOD Generalization (ICLR 2024 · similarity 0.84)
- Class Incremental Learning via Likelihood Ratio Based Task Prediction (ICLR 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
