Research map / Label-efficient and robust learning / Recognition and long-tail
H-Sets: Hessian-Guided Discovery of Set-Level Feature Interactions in Image Classifiers
Ayushi Mehrotra, Dipkamal Bhusal, Michael Clifford, Nidhi Rastogi
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 SAM segmentation prior (vs none) (ablation) | H-Sets (ResNet) · ImageNet | ROAD AOPC: 0.18 -> 0.26 | Improved |
| add SAM segmentation prior (vs none) (ablation) | H-Sets (DenseNet) · CUB | ROAD AOPC: 0.56 -> 0.65 | Improved |
| use SAM instead of Quickshift (ablation) | H-Sets (DenseNet) · CUB | ROAD AOPC: 0.58 -> 0.65 | Improved |
| add segmentation prior (SAM) to H-Sets (ablation) | H-Sets (DenseNet) · ImageNet | ROAD AOPC: 0.39 -> 0.38 | No clear effect |
| seed from top-IG pixel instead of random seed (with SAM) (ablation) | H-Sets (ResNet) · ImageNet | ROAD AOPC: 0.17 -> 0.26 | Improved |
| seed from top-IG pixel instead of random seed (with SAM) (ablation) | H-Sets (DenseNet) · CUB | ROAD AOPC: 0.52 -> 0.65 | Improved |
| use more interaction sets (1 to 5) (ablation) | H-Sets (ResNet) · CUB | ROAD AOPC: 0.58 -> 0.63 | Improved |
| use more interaction sets (1 to 9) (ablation) | H-Sets (VGG) · ImageNet | Gini sparsity: 0.98 -> 0.92 | Hurt |
| increase number of features nu from 250 to 5000 (ablation) | H-Sets · ImageNet | Gini sparsity: 0.99 -> 0.85 | Hurt |
| increase number of features nu from 250 to 5000 (ablation) | H-Sets · ImageNet | ROAD AOPC: 0.42 -> 0.37 | No clear effect |
| vary Hessian threshold mu from 0.1 to 0.8 (ablation) | H-Sets · ImageNet | ROAD AOPC: 0.39 -> 0.39 | No clear effect |
| H-Sets vs MoXI on faithfulness | H-Sets · ImageNet (ResNet) | ROAD AOPC: 0.19 -> 0.26 | Improved |
Closest papers by meaning
- Learn to Predict Sets Using Feed-Forward Neural Networks (TPAMI 2021 · similarity 0.85)
- Matching Feature Sets for Few-Shot Image Classification (CVPR 2022 · similarity 0.85)
- SetCon: Towards Open-Ended Referring Segmentation via Set-Level Concept Prediction (NeurIPS 2026 · similarity 0.84)
- Hierarchy-Agnostic Unsupervised Segmentation: Parsing Semantic Image Structure (NeurIPS 2024 · similarity 0.84)
- Learning Hierarchical Image Segmentation For Recognition and By Recognition (ICLR 2024 · similarity 0.84)
- DH-Set: Improving Vision-Language Alignment with Diverse and Hybrid Set-Embeddings Learning (CVPR 2025 · 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.
