Research map / Label-efficient and robust learning / OOD and anomaly detection
On the Out-of-distribution Generalization of Probabilistic Image Modelling
Mingtian Zhang, Andi Zhang, Steven McDonagh
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 |
|---|---|---|---|
| Use local model instead of full PixelCNN on in-distribution data | Local autoregressive · Fashion MNIST ID | BPD: 2.78 -> 2.89 | Hurt |
| Use local model instead of full PixelCNN on in-distribution data | Local autoregressive · CIFAR10 ID | BPD: 3.12 -> 3.25 | Hurt |
| Use local model instead of full PixelCNN on OOD data | Local autoregressive · Trained on Fashion MNIST, OOD MNIST/KMNIST | BPD: 1.50 -> 1.49 | No clear effect |
| Use local model instead of full PixelCNN on OOD data | Local autoregressive · Trained on CIFAR10, SVHN/CelebA | BPD: 2.13 -> 2.13 | No clear effect |
| Increase local horizon size h from 1 to 5 (ablation) | Local model · Fashion MNIST ID | BPD: 3.17 -> 2.88 | Improved |
| Increase local horizon size beyond h=2 (ablation) | Local model · Fashion MNIST trained, OOD MNIST | BPD: 1.48 -> 1.51 | Hurt |
| Increase local horizon size from h=2 to h=5 (CIFAR10) (ablation) | Local model · CIFAR10 ID | BPD: 3.38 -> 3.25 | Improved |
| Increase horizon size from h=2 to h=5 on OOD (CIFAR10 trained) (ablation) | Local model · SVHN | BPD: 2.21 -> 2.15 | No clear effect |
| Use non-local likelihood ratio score instead of full PixelCNN likelihood for OOD detection | Our method · CIFAR10 vs CelebA | AUROC: 0.602 -> 0.949 | Improved |
| Use non-local likelihood ratio score instead of full PixelCNN likelihood | Our method · FashionMNIST vs MNIST | AUROC: 0.074 -> 1.000 | Improved |
| Increase NeLLoC residual blocks r from 0 to 3 (trained on CIFAR) (ablation) | NeLLoC · CIFAR ID | BPD: 3.38 -> 3.25 | Improved |
| Split full ImageNet images into 400 patches for parallel decoding | NeLLoC · Full ImageNet | BPD: 3.25 -> 3.34 | Hurt |
Closest papers by meaning
- Understanding Failures in Out-of-Distribution Detection with Deep Generative Models (ICML 2021 · similarity 0.89)
- Understanding Likelihood of Normalizing Flow and Image Complexity through the Lens of Out-of-Distribution Detection (AAAI 2024 · similarity 0.88)
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 2024 · similarity 0.88)
- Leveraging Perturbation Robustness to Enhance Out-of-Distribution Detection (CVPR 2025 · similarity 0.87)
- Detecting Out-of-Distribution Through the Lens of Neural Collapse (CVPR 2025 · similarity 0.87)
- Towards Optimal Feature-Shaping Methods for Out-of-Distribution Detection (ICLR 2024 · similarity 0.87)
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.
