Training a Student Expert via Semi-Supervised Foundation Model Distillation

TL;DR

This paper presents a semi-supervised knowledge distillation framework that compresses large vision foundation models into smaller, efficient experts for instance segmentation, leveraging limited labeled data and extensive unlabeled data.

Contribution

The authors introduce a novel three-stage semi-supervised distillation method with an instance-aware contrastive loss for effective model compression and improved segmentation performance.

Findings

The student model achieves +11.9 AP over zero-shot teacher on Cityscapes.

The approach surpasses adapted teachers by +3.4 AP.

State-of-the-art results on benchmark datasets.

Abstract

Foundation models deliver strong perception but are often too computationally heavy to deploy, and adapting them typically requires costly annotations. We introduce a semi-supervised knowledge distillation (SSKD) framework that compresses pre-trained vision foundation models (VFMs) into compact experts using limited labeled and abundant unlabeled data, and instantiate it for instance segmentation where per-pixel labels are particularly expensive. The framework unfolds in three stages: (1) domain adaptation of the VFM(s) via self-training with contrastive calibration, (2) knowledge transfer through a unified multi-objective loss, and (3) student refinement to mitigate residual pseudo-label bias. Central to our approach is an instance-aware pixel-wise contrastive loss that fuses mask and class scores to extract informative negatives and enforce clear inter-instance margins. By maintaining this contrastive signal across both adaptation and distillation, we align teacher and student embeddings and more effectively leverage unlabeled images. On Cityscapes and ADE20K, our $\approx 11\times$ smaller student improves over its zero-shot VFM teacher(s) by +11.9 and +8.6 AP, surpasses adapted teacher(s) by +3.4 and +1.5 AP, and outperforms state-of-the-art SSKD methods on benchmarks.