Research map / Detection and segmentation / Open-set and few-shot detection
Exploring Why Object Recognition Performance Degrades Across Income Levels and Geographies with Factor Annotations
Laura Gustafson, Megan Richards, Melissa Hall, Caner Hazirbas, Diane Bouchacourt, Mark Ibrahim
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 texture-bias mitigation (Geirhos et al. training) to ResNet-50 | ResNet-50 ImageNet-1k · Dollar Street overall | top-5 accuracy: 32.2 -> 33.0 | Improved |
| add texture-bias mitigation to ResNet-50 | ResNet-50 ImageNet-1k · Dollar Street texture-marked images, low income | top-5 accuracy: 25.2 -> 29.3 | Improved |
| add texture-bias mitigation to ResNet-50 | ResNet-50 ImageNet-1k · Dollar Street texture-marked images, Africa | top-5 accuracy: 18.6 -> 21.6 | Improved |
| add texture-bias mitigation to ResNet-50 | ResNet-50 ImageNet-1k · Dollar Street texture-marked images, Asia | top-5 accuracy: 24.4 -> 25.0 | Improved |
| vary the income level of Dollar Street images (high vs low) | CLIP ViT-B/32 · Dollar Street | top-5 accuracy: 92.6 -> 66.9 | Hurt |
| switch CLIP image encoder from ViT-B/32 to ResNet50 | CLIP · Dollar Street low income | top-5 accuracy: 66.9 -> 73.4 | Improved |
| switch CLIP image encoder from ViT-B/32 to ResNet101 | CLIP · Dollar Street low income | top-5 accuracy: 66.9 -> 68.2 | Improved |
| use large-scale self-supervised pretraining (SEER) instead of supervised ResNet50 IN21k | image classifier · Dollar Street low income | top-5 accuracy: 58.4 -> 64.3 | Improved |
| replace ViT with MLP Mixer (both IN21k) | image classifier · Dollar Street low income | top-5 accuracy: 61.3 -> 61.9 | No clear effect |
| replace supervised ViT/ResNet by BEiT | image classifier · Dollar Street low income | top-5 accuracy: 61.3 -> 51.7 | Hurt |
| higher image resolution (DPI) as a predictor of accuracy | CLIP · Dollar Street | Pearson correlation with top-5 accuracy: -0.035 (p = 0.00005) | No clear effect |
| larger ImageNet-21K class count as a predictor of accuracy | BEiT · Dollar Street | Pearson correlation with top-5 accuracy: 0.222 (p < 0.0001) | No clear effect |
Closest papers by meaning
- Does Progress On Object Recognition Benchmarks Improve Generalization on Crowdsourced, Global Data? (ICLR 2024 · similarity 0.88)
- Contour Integration Underlies Human-Like Vision (ICML 2025 · similarity 0.86)
- ImageNet-X: Understanding Model Mistakes with Factor of Variation Annotations (ICLR 2023 · similarity 0.86)
- Object Detection Benchmarks are Incomplete: The Role of Label Errors and Annotation Uncertainty (NeurIPS 2026 · similarity 0.86)
- Directional Confusions Reveal Divergent Inductive Biases Through Rate-Distortion Geometry in Human and Machine Vision (NeurIPS 2026 · similarity 0.85)
- Two Effects, One Trigger: On the Modality Gap, Object Bias, and Information Imbalance in Contrastive Vision-Language Models (ICLR 2025 · similarity 0.84)
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.
