ODIN: Out-of-Distribution Image Detection Benchmark Results

ODIN is a method for detecting out-of-distribution images in neural networks without modifying pre-trained models. The description includes experimental results, such as reducing the false positive rate from 34.7% to 4.3% on DenseNet applied to CIFAR-10. The work is by Shiyu Liang from the University of Illinois Urbana-Champaign.

Use Cases

• Benchmarking out-of-distribution detection methods based on the described ODIN technique.
• Evaluating neural network robustness across diverse architectures and datasets as mentioned in the experiments.
• Improving model reliability by implementing temperature scaling and input perturbation strategies detailed in the description.

Strengths

• Method demonstrated a specific performance improvement, reducing false positive rate from 34.7% to 4.3% in one experiment.
• Compatibility with diverse network architectures and datasets is explicitly stated in the description.

Limitations

• Row count is unknown, which may limit suitability assessment.
• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.

Provenance

Source

University of Illinois Urbana-Champaign

Collection Method

Proposed research method (ODIN) with experimental evaluation.