Google Research open-sources subjective image classification framework
Understand this story
Google Research proposed and open-sourced the Agile Deliberation framework, which breaks the construction of subjective visual classifiers like "healthy food"—for which there is no single correct answer—into two stages: concept definition and concept iteration. A vision-language model actively selects images with ambiguous semantic boundaries and hands them to the user for judgment, then feeds the feedback into an automatic prompt optimization loop. Google Research says it has open-sourced the underlying code and algorithm on GitHub and provides an interactive Colab/Jupyter demo that walks through the entire process using "healthy food" as an example. MarkTechPost reports that the framework calls Gemini 2.5 Pro and Gemini 2.5 Flash via API, requiring no fine-tuning but requiring a Gemini API key; it uses gemini-2.5-pro and gemini-2.5-flash by default. On the data side, the default 2 shards total about 260 MB, and all 100 shards total about 13 GB.
AI overview based on source reports · Updated 23 hr ago
Coverage timeline
Follow the coverage to explore different perspectives.
- MarkTechPostGoogle AI Research Releases Agile Deliberation Framework for Gray-Area Image Classifiers
Google Research released Agile Deliberation, a human-in-the-loop framework that helps users turn vague concepts like 'healthy food' into structured written definitions that also serve as prompts for a Gemini 2.5 Flash VLM classifier. It finds borderline images by meaning rather than classifier confidence, using borderline queries, CLIP/FAISS retrieval, dictionary-learning clusters and a UCB bandit, and averaged 10.5 F1 points above zero-shot Gemini 2.5 Flash across 2 concepts and 18 live sessions. The code is Apache 2.0 on GitHub with a demo notebook, and the work appears in a CVPR 2026 paper.
- Google ResearchAgile Deliberation: How AI helps humans define the gray areas
Google Research introduces Agile Deliberation, a framework that treats building subjective AI classifiers as an iterative conversation, splitting the task into concept scoping (breaking a broad idea into sub-concepts and exclusions) and concept iteration (using vision-language models to surface borderline images that challenge the user's definitions). In 18 live 1.5-hour user sessions across the subjective domains of "Healthy Food" and "Pay-to-Play" content moderation, the approach produced more consistent, higher-performing classifiers than zero-shot, automated collaborator, and manual deliberation baselines while reducing user fatigue and anxiety. The code and algorithms are open-sourced on GitHub, with a Colab/Jupyter demo available.
Activity for this story
- Current comparable activity
- 9
- Peak comparable activity
- 17Nov 5 14:00
- Change in the past 24 hours
- +21%
Trends compare the same fully observed participants, which may cover fewer participants than current activity. Move the pointer or tap the chart to inspect hourly activity; use the left and right arrow keys with a keyboard.