Hidden-Shot: Towards One-Shot Task Generalization for Low-Level Vision Generalist Models

2026-07-01 · Source: Computer Vision and Pattern Recognition · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Computer Vision · Depth: Expert, quick

Summary

Hidden-Shot is an implicit prompt mechanism designed to enhance one-shot task generalization in low-level vision generalist models, addressing their unverified effectiveness in zero/few-shot scenarios beyond learned tasks. The method extracts implicit visual task-based information, employs a global task-aware textural prompt, and selectively merges this with in-task processing data to improve new task capabilities. Its design allows for direct, cost-effective injection with minimal alteration to the original generalist model's architecture. A new data-driven C/U assessment framework, covering 3C4U (3 conventional, 4 unconventional) and 3C7U (3 conventional, 7 unconventional) scenarios, was introduced to systematically evaluate generalization. Experiments across seven and ten datasets demonstrate Hidden-Shot's superior performance over state-of-the-art vision generalist models in one-shot new tasks, while maintaining consistent performance on existing tasks.

Key takeaway

For Computer Vision Engineers developing low-level vision generalist models, Hidden-Shot offers a promising approach to overcome limitations in one-shot task generalization. You should consider integrating implicit prompt mechanisms like Hidden-Shot to enhance your models' adaptability to new, unseen tasks without requiring significant architectural modifications. Additionally, adopting the C/U assessment framework can provide a systematic way to evaluate your models' generalization capabilities across diverse conventional and unconventional scenarios.

Key insights

Hidden-Shot enables one-shot task generalization for low-level vision generalist models via an implicit prompt mechanism.

Principles

• Implicit prompting enhances generalization.
• Selective information merging improves adaptation.
• Cost-effective injection minimizes model alteration.

Method

Hidden-Shot extracts implicit visual task-based information, utilizes a global task-aware textural prompt, and selectively merges this with in-task processing information to enhance one-shot capabilities in new tasks.

In practice

• Apply implicit prompts for new task adaptation.
• Use C/U assessment for generalization testing.
• Integrate direct injection for minimal model changes.

Topics

• Low-Level Vision
• Generalist Models
• One-Shot Learning
• Task Generalization
• Implicit Prompting
• Computer Vision
• C/U Assessment

Best for: Research Scientist, AI Scientist, Computer Vision Engineer, Machine Learning Engineer

Related on AIssential

• Beyond Low-Rank: Low-Rank Sparse Prompting via Spiking Neural Network and Prompt Factorization — LoRSP uses spiking neural networks and low-rank factorization to create sparse, efficient visual prompts for vision models.
• One-Shot Novel View and Pose Human Image Synthesis via 3D Prior Guided Diffusion Model — A 3D prior-guided conditional diffusion model enables high-quality one-shot novel view and pose human image synthesis.
• MaskWAM: Unifying Mask Prompting and Prediction for World-Action Models — MaskWAM unifies mask prompting and prediction via a Mixture of Transformers to enhance robotic control by resolving spatial ambiguity and visual noise.
• Domain-Guided Prompting of the Segment Anything Model for Seismic Interpretation: The Role of Attributes, Visualization, and Hybrid Prompts — Zero-shot SAM adaptation for seismic interpretation is achieved by combining domain-guided visualization with hybrid prompting.
• Take a Peek: Efficient Encoder Adaptation for Few-Shot Semantic Segmentation via LoRA — Take a Peek (TaP) efficiently adapts FSS encoders to novel classes using Low-Rank Adaptation (LoRA) on support sets.
• Temporal Stability and Few-Shot Prompting in Math Task Assessment — Prompt engineering offers more reliable AI performance improvements on specialized tasks than passive model version updates.

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by Computer Vision and Pattern Recognition.