Visual Prompt Learning
Overview
Visual prompt learning studies how to steer pretrained models toward new tasks by changing prompts rather than extensively retraining the models. Approaches include soft tokens inserted into model layers, prompt pools with dynamic routing, and visual cues such as masks, points, overlays, or selected frames; some methods also align visual prompts with text. These techniques are used for tasks including classification, segmentation, detection, and multimodal reasoning, often with frozen backbones and little task-specific data. A central challenge is adapting to new domains and noisy inputs while preserving the pretrained model’s generalization and keeping prompts efficient.
Latest papers
129 papers7 papers in the last four weeks, down 12% on the four weeks before. 0.1% of all new papers.