SpikingResNet

Out-of-distribution (OoD) detectors, built on top of models trained on a single dataset (in-distribution, ID), should flag test-time inputs that don't belong to that distribution. Spike-like models replace standard ReLU activations with leaky integrate-and-fire (LIF) neurons on conventional ResNet backbones, operating at a single inference time step to achieve binary spike activations and substantially reduced arithmetic complexity, while remaining competitive with full-precision CNN baselines on OpenOOD benchmarks.

Full code: https://github.com/Alex-TLR/spikingResNet

About the OoD experiments

Two difficulties are evaluated: near-OoD (a semantically similar dataset, e.g. CIFAR-100 against a CIFAR-10 model) and far-OoD (a very different one, e.g. SVHN/MNIST). Detectors are scored by AUROC (higher is better) and FPR95 (false-positive rate at 95% true-positive rate, lower is better). The experiments conform to the OpenOoD benchmark.

How this repository is organized

The code includes the inference-only pipeline, the OoD benchmarks, and one folder per model "family" (architecture + dataset). Within each family, checkpoints are grouped by experiment (loss function + training time steps), then by expansion/seed:

spikingresnet/
├── code/                                   <- inference and OoD code
├── checkpoints.csv                         <- every checkpoint in this repo, one row each
└── <family>/<experiment>/T<T>_E<E>_S<seed>/
    ├── config.json
    └── model.safetensors

495 checkpoints across 7 families:

• resnet10-cifar10 -- CIFAR10, 25 checkpoints
• resnet18-cifar10 -- CIFAR10, 25 checkpoints
• resnet4-cifar10 -- CIFAR10, 25 checkpoints
• spikingresnet4-cifar10 -- CIFAR10, 75 checkpoints
• spikingresnet10-cifar10 -- CIFAR10, 75 checkpoints
• spikingresnet18-cifar10 -- CIFAR10, 150 checkpoints
• spikingresnet18-cifar100 -- CIFAR100, 120 checkpoints

Every experiment sweeps all 5 seeds (42, 1987, 1991, 2020, 2024) across its listed expansions -- checkpoints.csv gives the exact seed of every checkpoint -- see "Running an OoD benchmark" below to reproduce the results.

Quick start

Install the required packages:

pip install -r code/requirements.txt

See what models are available:

python code/run_example.py dsluga/spikingresnet --list
python code/run_example.py dsluga/spikingresnet --list resnet10-cifar10

Run one checkpoint:

python code/run_example.py dsluga/spikingresnet --subfolder resnet10-cifar10/cross-entropy-baseline/T1_E10_S1987

Running an OoD benchmark

The option --experiment picks which detector to run:

• posthoc (default) -- scoring computed directly from a trained model's outputs, no extra training or feature bank needed: MSP (max softmax probability), Energy, MLS (max logit score), and ASH (activation shaping);
• knn -- a k-nearest-neighbors detector over penultimate-layer features.

Sample run of the knn OoD detection benchmark:

python code/run_example.py --experiment knn dsluga/spikingresnet --all-seeds \
    --subfolder resnet10-cifar10/cross-entropy-baseline/T1_E10

Datasets

By default, the benchmark only uses datasets that torchvision downloads automatically (CIFAR-10/100, SVHN, MNIST). The full benchmark also evaluates on Textures (DTD), Places365, and Tiny-ImageNet-200; fetch those directly from their original hosts with:

python code/spikingresnet_code/download_datasets.py textures tiny_imagenet   # ~850MB
python code/spikingresnet_code/download_datasets.py places365                # ~25-30GB

then edit code/run_example.py's NEAR_OOD/FAR_OOD lists to include them in a benchmark run, e.g.:

NEAR_OOD = ['CIFAR10', 'CIFAR100', 'tImage200']
FAR_OOD =  ['MNIST', 'SVHN', 'Textures', 'Places365']

Model list

resnet10-cifar10

Dataset: CIFAR10

25 checkpoints total. The table above groups them by experiment; every individual subfolder/config/seed is listed in the repo-root checkpoints.csv.

Example: subfolder="resnet10-cifar10/cross-entropy-baseline/T1_E10_S1987"

resnet18-cifar10

Dataset: CIFAR10

25 checkpoints total. The table above groups them by experiment; every individual subfolder/config/seed is listed in the repo-root checkpoints.csv.

Example: subfolder="resnet18-cifar10/cross-entropy-baseline/T1_E1_S42"

resnet4-cifar10

Dataset: CIFAR10

25 checkpoints total. The table above groups them by experiment; every individual subfolder/config/seed is listed in the repo-root checkpoints.csv.

Example: subfolder="resnet4-cifar10/cross-entropy-baseline/T1_E10_S1987"

spikingresnet4-cifar10

Dataset: CIFAR10

75 checkpoints total. The table above groups them by experiment; every individual subfolder/config/seed is listed in the repo-root checkpoints.csv.

Example: subfolder="spikingresnet4-cifar10/mse-loss-t1-train/T1_E10_S1987"

spikingresnet10-cifar10

Dataset: CIFAR10

75 checkpoints total. The table above groups them by experiment; every individual subfolder/config/seed is listed in the repo-root checkpoints.csv.

Example: subfolder="spikingresnet10-cifar10/mse-loss-t1-train/T1_E10_S1987"

spikingresnet18-cifar10

Dataset: CIFAR10

150 checkpoints total. The table above groups them by experiment; every individual subfolder/config/seed is listed in the repo-root checkpoints.csv.

Example: subfolder="spikingresnet18-cifar10/mse-loss-t1-train/T1_E5_S42"

spikingresnet18-cifar100

Dataset: CIFAR100

120 checkpoints total. The table above groups them by experiment; every individual subfolder/config/seed is listed in the repo-root checkpoints.csv.

Example: subfolder="spikingresnet18-cifar100/mse-loss-t1-train/T1_E50_S42"

Citation

@article{spikingresnet_ood,
  title   = {Out-of-Distribution Detection with Spike-Like Networks: Population Coding, Training Depth, and Single-Step Feature Representations},
  author  = {A. Avramović, S. Gajić, V. Jovanović, V. Risojević, D. Sluga},
  journal = {},
}