SpikeCLR: Contrastive Self-Supervised Learning for Few-Shot Event-Based Vision using Spiking Neural Networks

Maxime Vaillant; Axel Carlier; Lai Xing Ng; Christophe Hurter; Benoit R. Cottereau

cs.CVMar 17, 2026

SpikeCLR: Contrastive Self-Supervised Learning for Few-Shot Event-Based Vision using Spiking Neural Networks

Authors: Maxime Vaillant, Axel Carlier, Lai Xing Ng, Christophe Hurter, Benoit R. Cottereau

Organizations: CNRS, IPAL IRL 2955, Singapore · Université de Toulouse, IRIT, France · Institute for Infocomm Research, A*STAR, Singapore · Fédération ENAC ISAE-SUPAERO ONERA, Université de Toulouse, France ·

Abstract

Event-based vision sensors provide significant advantages for high-speed perception, including microsecond temporal resolution, high dynamic range, and low power consumption. When combined with Spiking Neural Networks (SNNs), they can be deployed on neuromorphic hardware, enabling energy-efficient applications on embedded systems. However, this potential is severely limited by the scarcity of large-scale labeled datasets required to effectively train such models. In this work, we introduce SpikeCLR, a contrastive self-supervised learning framework that enables SNNs to learn robust visual representations from unlabeled event data. We adapt prior frame-based methods to the spiking domain using surrogate gradient training and introduce a suite of event-specific augmentations that leverage spatial, temporal, and polarity transformations. Through extensive experiments on CIFAR10-DVS, N-Caltech101, N-MNIST, and DVS-Gesture benchmarks, we demonstrate that self-supervised pretraining with subsequent fine-tuning outperforms supervised learning in low-data regimes, achieving consistent gains in few-shot and semi-supervised settings. Our ablation studies reveal that combining spatial and temporal augmentations is critical for learning effective spatio-temporal invariances in event data. We further show that learned representations transfer across datasets, contributing to efforts for powerful event-based models in label-scarce settings.

Figures & tables

Panel of Figure 1 — Figure 1: Effect of each augmentation family on an event histogram from CIFAR10-DVS. (a) Original sample. (b) Spatial: rolling shifts the content spatially. (c) Polarity: brightness scaling modifies event intensities. (d) Temporal: cropping a different time window changes the motion phase. More examples in Supplementary Material.

Panel of Figure 2 — Figure 2: Classification accuracy on CIFAR10-DVS. (a) Few-shot comparison across methods (averaged across 10 random subsets). (b) Accuracy as a function of the percentage of labeled training data (averaged across 3 random subsets).

Figure 3: Few-shot accuracy on CIFAR10-DVS across different SNN backbones (averaged across 3 random subsets). † denote separable convolution.

Augmentation Set	1-shot	9-shot	45-shot
EventDrop [ 19 ]	18.4 $(\pm 2.5)$	30.1 $(\pm 0.5)$	42.5 $(\pm 0.8)$
NDA † [ 26 ]	16.3 $(\pm 2.9)$	28.9 $(\pm 0.4)$	41.3 $(\pm 2.2)$
Spatial only	17.4 $(\pm 2.0)$	27.7 $(\pm 0.9)$	39.9 $(\pm 1.2)$
Temporal only	18.1 $(\pm 2.7)$	32.1 $(\pm 0.7)$	44.3 $(\pm 1.0)$
Polarity only	17.4 $(\pm 1.5)$	30.2 $(\pm 2.2)$	41.8 $(\pm 2.2)$
Spatial + Temporal	20.3 $(\pm 2.7)$	34.7 $(\pm 0.3)$	47.1 $(\pm 1.6)$

Table 1: Ablation study of augmentation families on CIFAR10-DVS. Few-shot fine-tuning accuracy (%) reported as mean

\pm

std over 3 random subsets.

Loss	1-shot	9-shot	45-shot
NT-Xent	20.2 $(\pm 2.8)$	35.2 $(\pm 0.7)$	48.3 $(\pm 1.7)$
Temporal NT-Xent	21.2 $(\pm 3.8)$	37.3 $(\pm 1.6)$	49.4 $(\pm 1.1)$
Supervised	11.9 $(\pm 1.3)$	26.0 $(\pm 0.2)$	38.8 $(\pm 0.8)$

Table 2: Ablation study of contrastive loss computation on CIFAR10-DVS. Few-shot fine-tuning accuracy (%) reported as mean

\pm

std over 3 random subsets.

Panel of Figure 4 — Figure 4: Classification accuracy on DVS-Gesture. (a) Few-shot comparison across methods (mean and std over 10 random subsets). (b) Accuracy as a function of the percentage of labeled training data (mean and std over 3 random subsets).

Dataset		1-shot	9-shot	45-shot
Pretrain	Target
CIFAR10-DVS	CIFAR10-DVS	20.2 $(\pm 2.8)$	35.2 $(\pm 0.7)$	48.3 $(\pm 1.7)$
N-Caltech101	CIFAR10-DVS	18.6 $(\pm 1.7)$	31.2 $(\pm 1.4)$	42.6 $(\pm 0.9)$
-	CIFAR10-DVS	11.9 $(\pm 1.3)$	26.0 $(\pm 0.2)$	38.8 $(\pm 0.8)$
DVS-Gesture	DVS-Gesture	34.2 $(\pm 2.2)$	65.8 $(\pm 2.9)$	84.3 $(\pm 2.4)$
CIFAR10-DVS	DVS-Gesture	35.1 $(\pm 2.4)$	67.4 $(\pm 4.5)$	84.5 $(\pm 1.0)$

Table 3: Cross-dataset transfer learning results. Few-shot fine-tuning accuracy (%) reported as mean

\pm

std over 3 random subsets. Models pretrained on different event datasets are fine-tuned on distinct target datasets and evaluated in few-shot scenarios.

Amount of pretrained data	1-shot	10-shot	50-shot
10K	56.9 $(\pm 7.3)$	91.6 $(\pm 0.8)$	96.6 $(\pm 0.3)$
60K	61.3 $(\pm 7.8)$	94.4 $(\pm 0.4)$	97.5 $(\pm 0.3)$
Supervised	21.3 $(\pm 5.0)$	51.7 $(\pm 4.9)$	95.2 $(\pm 0.3)$

Table 4: Impact of pretraining dataset size on N-MNIST. Few-shot fine-tuning accuracy (%) reported as mean

\pm

std over 10 random subsets. Models pretrained with different amounts of unlabeled data are compared against supervised baselines.

Self-Supervised Visual Representation Learning

Self-supervised visual representation learning trains image, video, and 3D encoders on unlabeled data to produce transferable visual features.

See all 235 papers

SICAF: Time-Varying Focus Bottleneck for Self-Supervised Event-Based Optical Flow with Spiking Neural Network
Oct 21, 2026Shuangming Yang, Qing He, Shangqi Guo +1Event-Based Optical Flow Spiking Neural Networks
Tianjin University, Tianjin, China · Tsinghua University, Beijing, China · Xi’an Jiaotong University, Xi’an, China
Toward On-Chip Training of Spiking Neural Networks for Dense Event-Based Vision
Sep 26, 2026Maxime Vaillant, Axel Carlier, Lai Xing Ng +2Spiking Neural Networks Local Learning
CNRS, IPAL IRL 2955, Singapore · Université de Toulouse, IRIT, France · Institute for Infocomm Research, A*STAR, Singapore +2
Sequence-SOD: Bio-inspired Sequence-aware Spiking ObjectDetection for Event Cameras
Jul 29, 2026Katharina Bendig, René Schuster, Didier StrickerSpiking Neural Networks Event-Based Object Detection
RPTU – Technical University Kaiserslautern-Landau, Gottlieb-Daimler-Straße 47, Kaiserslautern 67663, Germany · DFKI – German Research Center for Artificial Intelligence, Trippstadter Str. 122, Kaiserslautern 67663, Germany

Oct 21, 2026J/K move · Enter open · S save

SICAF: Time-Varying Focus Bottleneck for Self-Supervised Event-Based Optical Flow with Spiking Neural Network

Shuangming Yang, Qing He, Shangqi Guo, Badong Chen

Tianjin University, Tianjin, China · Tsinghua University, Beijing, China · Xi’an Jiaotong University, Xi’an, China

Event-Based Optical Flow Spiking Neural Networks Information Bottleneck

Open paper AskPDF arXiv

Event-based optical flow estimation achieves high temporal resolution and dynamic range by exploiting asynchronous events, while self-supervised learning provides an effective solution to the absence of ground-truth labels. Spiking neural networks (SNNs), as brain-inspired models, process information through asynchronous spikes, naturally aligning with optical flow events. However, the temporal irregularity across event streams hinders the compact extraction of spatiotemporal features, while the absence of explicit labels leads to spurious correlations, thereby increasing redundancy within latent representations. In this paper, we propose a novel 'Compactly Compress, Time-varying Focus' training strategy. We present SICAF (Spikes Information Compression And Focus) framework that formulates the learning process as a constrained optimization with temporal modulation. It encourages self-supervised SNNs to extract compact latent representations while preventing over-compression that would discard temporal saliency. To better enable feature selectivity across time steps, we design an estimator with learnable temporal attention, achieving context awareness with cross-time dependencies. Experimental results on several datasets demonstrate that SICAF achieves state-of-the-art performance in self-supervised SNNs, with a 8.14% improvement in AEE on the MVSEC dataset and a 13.49% increase in robustness under white-box adversarial attacks.

Abstract

Self-Supervised Visual Representation Learning

Explore similar work

SICAF: Time-Varying Focus Bottleneck for Self-Supervised Event-Based Optical Flow with Spiking Neural Network

Toward On-Chip Training of Spiking Neural Networks for Dense Event-Based Vision

Sequence-SOD: Bio-inspired Sequence-aware Spiking ObjectDetection for Event Cameras