Research map / Adversarial robustness and security / Certified and empirical robustness
Contemplating Real-World Object Classification
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Classify the isolated object bounding box instead of the full scene | ImageNet-pretrained classification models · ObjectNet, 113 ImageNet-overlap categories | accuracy: around 20-30% performance improvement | Helped |
| Classify the isolated object bounding box instead of the full scene | VGG-19, Inception-v3, and ResNet-152 · ObjectNet, 14 natural image distortions | top-1 and top-5 accuracy under corruption | Helped |
| Classify the isolated object bounding box instead of the full scene | VGG-19, Inception-v3, and ResNet-152 · ObjectNet, FGSM | accuracy under attack: 0.53/2.48 -> 3.27/10.44; 3.18/10.62 -> 9.03/25.13; 2.39/10.34 -> 10.62/26.64 | Helped |
| Train and test on segmented objects instead of full images | ResNet-18 finetuned on ImageNet · MS COCO, 10 classes | test accuracy: 66.9% -> 80.3% | Helped |
| Train and test on segmented objects instead of bounding boxes | ResNet-18 finetuned on ImageNet · MS COCO, 10 classes | test accuracy: 78% -> 80.3% | Helped |
| Use segmentation masks instead of full images or bounding boxes | ResNet-18 finetuned on ImageNet · MS COCO, 10 classes, FGSM and PGD-5 | accuracy under adversarial attack | Helped |
| Mask perturbations that fall on the background | ResNet-18 trained on segmented objects · MS COCO, 10 classes, FGSM and PGD-5 | accuracy under adversarial attack | Helped |
| Use segmentation masks instead of full images or bounding boxes | ResNet-18 finetuned on ImageNet · MS COCO, 10 classes, 14 noise distortions | accuracy under corruption | Helped |
| Train on segmented objects rather than full images or bounding boxes | ResNet-18 finetuned on ImageNet · MS COCO, 10 classes, scale, in-plane rotation, and horizontal translation | top-1 accuracy under geometric transformations | Helped |
| Remove the background from noisy MNIST and FashionMNIST images | Two-layer CNNs · MNIST and FashionMNIST, FGSM | accuracy under attack | Helped |
Closest papers by meaning
- ObjectNet: A large-scale bias-controlled dataset for pushing the limits of object recognition models (NeurIPS 2019 · similarity 0.91)
- Progress and Limitations of Deep Networks to Recognize Objects in Unusual Poses (AAAI 2023 · similarity 0.91)
- ImageNet-E: Benchmarking Neural Network Robustness via Attribute Editing (CVPR 2023 · similarity 0.89)
- Intriguing Properties of Generative Classifiers (ICLR 2024 · similarity 0.87)
- Doomed to Re-Annotate, Forever: The ImageNet Story (NeurIPS 2026 · similarity 0.86)
- How to Evaluate the Generalization of Detection? A Benchmark for Comprehensive Open-Vocabulary Detection (AAAI 2024 · similarity 0.86)
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.