Ranked Activation Shift for Out-of-Distribution Detection 📈

This is the source code for the paper "Ranked Activation Shift for Out-of-Distribution Detection" by Gianluca Guglielmo and Marc Masana.

The paper introduces RAS, a novel score-enhancing OoD detection method that is easy to use, hyperparameter-free and requires no assumptions on the model architecture.

Visual intuition of RAS.

---

🎯 Motivation

In many real-world applications, neural networks may encounter data that do not belong to the distribution on which they were trained. Detecting such OoD samples is critical to prevent unsafe decisions in areas such as autonomous driving, healthcare, or finance.

---

📌 Overview

This toy example uses:

• CIFAR-10, CIFAR-100, ImageNet200, ImageNet as in-distribution (ID) datasets.
• ResNet18 (with pretrained weights) as the backbone.

The script stores feature vectors from each sample and applies RAS to them. It then computes the AUROC score per layer.

After running the script, you will obtain a bar plot with the AUROC calculated for each layer of the model. The dotted line indicates the AUROC of the final layer. Green layers are the ones that beat the final layer.

---

🚀 How to Run

📦 Requirements

• Python 3.10+
• CUDA-capable GPU (recommended)

Install the dependencies:

pip install -r requirements.txt

📂 Datasets

CIFAR-10, CIFAR-100, and ImageNet-200 datasets (along with their OoD counterparts) are downloaded automatically by the OpenOOD benchmark when you first run the script. No manual setup is needed.

ImageNet-1K requires manual preparation because the Google Drive download from OpenOOD may fail due to file size and quota limits. To set it up:

1. Download the OpenOOD imagenet_1k archive from Google Drive (you may need to request access).
2. Extract it into data/images_largescale/imagenet_1k/.
3. You also need the OoD datasets for the ImageNet benchmark (ssb_hard, ninco, inaturalist, texture, openimage_o, imagenet_v2, imagenet_c, imagenet_r). These are downloaded automatically if not already present.

The expected directory structure after setup:

data/
├── benchmark_imglist/       # auto-downloaded
├── images_classic/          # auto-downloaded (CIFAR, SVHN, MNIST, ...)
│   ├── cifar10/
│   ├── cifar100/
│   ├── ...
└── images_largescale/       # auto-downloaded (except imagenet_1k)
    ├── imagenet_1k/
    ├── ssb_hard/
    ├── ...

▶ Running the Script

Run all experiments (CIFAR-10, CIFAR-100, ImageNet-200, ImageNet):

python run_ras.py

Run a single dataset:

python run_ras.py --id-data cifar10
python run_ras.py --id-data cifar100
python run_ras.py --id-data imagenet200
python run_ras.py --id-data imagenet

Optional arguments:

Argument	Default	Description
--id-data	all	Which ID dataset to evaluate (cifar10, cifar100, imagenet200, imagenet, or all)
--batch-size	200	Batch size for data loaders

Results are saved as pickle files in the run_results/ directory.

---

📚 References

COMING SOON