aiqbench analysis
This research introduces SPARC, a self-supervised learning framework that enhances dense prediction tasks by incorporating region-level contrastive objectives alongside traditional image-level goals. By leveraging superpixels to establish feature correspondences, the method significantly outperforms existing approaches like MoCo-v2 and DenseCL in semantic segmentation and object detection without requiring manual annotations. The study validates that focusing on regional representations is a critical advancement for improving visual pretraining efficiency in computer vision applications.
SPARC: SuperPixel-Aware Region Contrastive Learning for Self-Supervised Dense Prediction
arXiv:2609.25067v1 Announce Type: cross Abstract: Self-supervised learning (SSL) has become an effective approach for learning visual representations without manual annotations. Among SSL approaches, contrastive learning has been widely used for visual representation learning. However, existing contrastive SSL methods have focused primarily on image-level or pixel-level representation learning, while region-level representation learning remains less explored. We propose SPARC, a region-level contrastive learning framework that leverages superpixels to establish explicit correspondence between augmented image views. SPARC introduces a region contrastive branch that performs superpixel-based feature pooling and optimizes a region-level contrastive objective jointly with a global image-level objective. Under identical settings, SPARC consistently outperforms previous methods such as MoCo-v2 and DenseCL, achieving improvements of up to +9.79 mIoU for semantic segmentation and +4.88 AP for object detection. Ablation studies further demonstrate that region-level objectives produce the strongest performance. Thus, region-level contrastive learning is an effective approach for improving self-supervised visual pretraining for dense prediction tasks. Code repository can be accessed at https://github.com/xRIPEIx/SPARC.
Abstract:Self-supervised learning (SSL) has become an effective approach for learning visual representations without manual annotations. Among SSL approaches, contrastive learning has been widely used for visual representation learning. However, existing contrastive SSL methods have focused primarily on image-level or pixel-level representation learning, while region-level representation learning remains less explored. We propose SPARC, a region-level contrastive learning framework that leverages superpixels to establish explicit correspondence between augmented image views. SPARC introduces a region contrastive branch that performs superpixel-based feature pooling and optimizes a region-level contrastive objective jointly with a global image-level objective. Under identical settings, SPARC consistently outperforms previous methods such as MoCo-v2 and DenseCL, achieving improvements of up to +9.79 mIoU for semantic segmentation and +4.88 AP for object detection. Ablation studies further demonstrate that region-level objectives produce the strongest performance. Thus, region-level contrastive learning is an effective approach for improving self-supervised visual pretraining for dense prediction tasks. Code repository can be accessed at this https URL.
5 pages, 2 figures. Submitted to the IEEE ICASSP 2027 Conference
Computer Vision and Pattern Recognition (cs.CV); Machine Learning (cs.LG)
arXiv:2609.25067 [cs.CV]
(or arXiv:2609.25067v1 [cs.CV] for this version)
https://doi.org/10.48550/arXiv.2609.25067 Focus to learn more arXiv-issued DOI via DataCite
arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.
Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.
Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.
Full text under the publisher's open licence · 266 words
Related coverage
Masked Generative-Contrastive Representation Learning for Cross-Dataset EEG-Based Emotion Recognition
arXiv · Sep 9, 2026
Improving CoT Monitorability of Evaluation Awareness via Verbalization Training
LessWrong · Oct 2, 2026
Masked Topology Modeling for Self-Supervised Learning on Parametric CAD
arXiv · Jul 24, 2026
Component-Aware Structure-Preserving Style Transfer for Satellite Visual Sim2Real Data Construction
arXiv · Aug 24, 2026