Zero-Shot Classification & AI Grouping

Traditional photo management requires manually creating folders and labels, which is nearly impossible with massive photo libraries. Photocatalyst's zero-shot classification lets you define custom category names, and AI automatically classifies the entire library — no training samples, no manual labeling required.

1. What is Zero-Shot Classification?

Zero-shot classification is an AI classification technique that requires no training data. You simply provide category names and descriptions, and the AI understands the category meaning, automatically assigning photos to the corresponding categories.

Unlike traditional machine learning classification:

• No training samples: No need to provide example photos in advance
• Flexible expansion: Add new categories anytime, effective immediately
• Language understanding: AI matches photo content through semantic understanding

2. Creating Classifications

Entry Point

Sidebar → AI Grouping → Zero-Shot Classification

Adding a Single Category

1. Click the add button
2. Enter the category name (Chinese or English)
3. (Recommended) Fill in the description — the more specific the description, the more accurate the classification
4. Confirm to add

Description Writing Tips

The description is key to classification accuracy. The AI performs semantic matching between the description and photos:

Using English for descriptions is recommended for more accurate classification results (AI models have more mature semantic understanding of English).

Batch Adding Categories

1. Click the batch add button
2. Enter multiple category names, separated by commas, semicolons, or line breaks
3. Click "Default Categories" or "Default Categories (Simple)" to fill in preset category lists with one click
4. Confirm to add

Preset Default Categories

The system includes common classification themes; click to quickly fill them in: Landscape, Food, Pets, Architecture, People, Sports, Transportation, Plants, Night Scenes, Documents, and more.

3. Running Classification

Steps

1. Ensure at least two categories have been created
2. Click the "Run Classification" button
3. Choose a grouping mode:
   • Fixed Grouping: Each photo is assigned to the best-matching category
   • Search Grouping: Search for matching photos by category (a photo can belong to multiple categories)
4. Wait for AI processing to complete
5. View photos under each classification

Classification Result Management

• Photo count displayed per category
• Click to view all photos in that category
• Supports further filtering and batch operations

4. Grouping by Image

In addition to defining category-based classification, you can also use existing photos as references to automatically find similar photos.

Grouping by Similarity

Find similar images based on visual features (color, texture, composition, etc.):

1. Select one or more reference photos in the photo list
2. Right-click → Group by Image → Similarity Grouping
3. Adjust the similarity threshold (recommended starting from 70%)
4. Confirm and wait for analysis to complete

Grouping by Semantics

Find similar images based on AI semantic understanding (the subject matter expressed by the photo):

1. Select reference photos
2. Right-click → Group by Image → Semantic Grouping
3. The system returns groups of photos related by subject matter

Use Cases

5. OCR Text Grouping

Automatically recognize text in photos (such as screenshots, street signs, menus, etc.) and group by keywords.

1. AI Grouping page → OCR Grouping
2. Run OCR recognition and grouping
3. Browse by recognized text keywords

Suitable for: Screenshot organization, document scan classification, sign/menu photo management.

6. Folder Album (AI Folder Auto-Classification)

Use AI to automatically classify photos within a specific folder:

1. Navigate to the target folder in the folder tree
2. Click "Folder Album Classification"
3. The system uses zero-shot classification technology to automatically group
4. Generate multiple sub-albums

Ideal for quickly splitting a specific project folder into sub-themes.

Practical Example: Landscape Photographer Classification Scheme

Category System:
  Natural Landscapes
    ├── Mountains (mountain peaks, alpine scenery, summit views)
    ├── Waters (lake, river, ocean, waterfall, sea)
    ├── Forests (forest, woodland, trees, jungle)
    ├── Deserts (desert, sand dunes, arid landscape)
    └── Skies (clouds, sky, starry night, aurora, milky way)

  Cities & Architecture
    ├── City Skylines (city skyline, urban landscape)
    ├── Architectural Details (architectural detail, building texture)
    └── Interior Spaces (interior design, room, hall, church interior)

  Documentary & Culture
    ├── Street Photography (street photography, candid, urban life)
    ├── Festivals & Celebrations (festival, celebration, ceremony, parade)
    └── Markets & Bazaars (market, bazaar, vendor, food stall)

  Special Techniques
    ├── Long Exposure (long exposure, light trails, motion blur)
    ├── Macro (macro, close-up, insect, flower detail)
    └── Black & White (black and white, monochrome, high contrast)

Configure once, auto-classify the entire library.

Related sections: Chapter 5: AI Smart Features | Chapter 6: Multi-Dimensional Browsing