SAM GUI Tool - Interactive Image Segmentation Interface

A user-friendly graphical interface for the Segment Anything Model (SAM) that enables interactive image segmentation with real-time visualization and multiple operation modes.

🚀 Quick Start

1. Launch the Application

python test_gui.py

2. Load a SAM Model

• Click "选择模型文件夹" (Choose Model Folder)
• Navigate to your SAM model weights directory
• The tool will automatically detect and match model files:
  • sam_vit_b_*.pth → ViT-B model
  • sam_vit_l_*.pth → ViT-L model
  • sam_vit_h_*.pth → ViT-H model
• Select your preferred model architecture and click "加载模型" (Load Model)

3. Load an Image

• Click "加载图像" (Load Image)
• Select your target image (supports JPG, PNG, BMP, TIFF)
• The image will appear in the main canvas

4. Start Segmenting

Choose from three operation modes and start segmenting!

🎯 Operation Modes

📍 Point Mode (点击模式)

Best for: Precise object segmentation with user guidance

How to use:

1. Select "点击模式" (Point Mode)
2. Choose point type:
   • 前景点 (Foreground): Click on the object you want to segment
   • 背景点 (Background): Click on areas you want to exclude
3. Click on the image to add points
4. Click "生成mask" (Generate Mask) to create segmentation

Tips:

• Red circles = Foreground points
• Blue circles = Background points
• Use "撤销" (Undo) to remove the last point
• Multiple points improve accuracy

📦 Box Mode (框选模式)

Best for: Quick segmentation of rectangular regions

How to use:

1. Select "框选模式" (Box Mode)
2. Click and drag to draw a bounding box around your target object
3. Release mouse button - the mask will be generated automatically

Tips:

• Yellow rectangle shows your selection
• Drag from corner to corner for best results
• The mask will be generated immediately upon release

🤖 Auto Mode (自动分割)

Best for: Discovering all objects in an image

How to use:

1. Select "自动分割" (Auto Mode)
2. Click "生成mask" (Generate Mask)
3. Wait for processing (may take 10-30 seconds)
4. The largest detected object will be displayed

Tips:

• Processes the entire image automatically
• Shows the largest segmented region by default
• Great for exploration and object discovery

🎛️ Interface Overview

Model Configuration Panel (模型配置)

• Model Architecture: Choose between ViT-B (fastest), ViT-L (balanced), ViT-H (best quality)
• Model Files: Automatic detection and loading of SAM weights
• Status Indicator: Real-time model loading status with color codes:
  • 🟠 Orange: Loading/Ready to load
  • 🟢 Green: Successfully loaded
  • 🔴 Red: Error occurred

File Operations (文件操作)

• 加载图像 (Load Image): Import images for segmentation
• 保存mask (Save Mask): Export binary mask as PNG
• 保存叠加图 (Save Overlay): Export image with colored mask overlay

Action Controls (操作)

• 清除 (Clear): Remove all annotations and masks
• 生成mask (Generate Mask): Create segmentation based on current inputs
• 撤销 (Undo): Remove the last point in Point Mode

Real-time Log Panel (日志信息)

• Live Logging: View detailed operation progress
• Auto-scroll: Automatically scroll to latest messages
• Log Controls:
  • 清空日志 (Clear Log): Remove all log entries
  • 保存日志 (Save Log): Export log to text file

🎨 Visual Feedback

Mask Visualization

• Green overlay: Segmented regions (70% image + 30% green)
• Real-time preview: Instant visual feedback during interaction

Point Annotations

• Red circles: Foreground points (include in mask)
• Blue circles: Background points (exclude from mask)
• White borders: Enhanced visibility on any background

Bounding Box

• Yellow rectangle: Current selection area in Box Mode

💡 Tips for Best Results

Point Mode Strategy

• Start with 1-2 foreground points on your target object
• Add background points near object boundaries for precision
• Use multiple foreground points for complex shapes
• Background points help separate similar-colored objects

Image Quality Tips

• Use high-resolution images when possible
• Ensure good contrast between objects
• Avoid heavily compressed or blurry images

Performance Optimization

• ViT-B: Choose for real-time interaction (fastest)
• ViT-L: Good balance of speed and quality
• ViT-H: Best quality for final results (slower)

🔧 Model Status Indicators

Status	Color	Meaning
检查模块中...	🟠 Orange	Checking SAM modules
模块已加载	🟠 Orange	Modules loaded, ready for model
已加载	🟢 Green	Model successfully loaded
加载失败	🔴 Red	Model loading failed

📁 Supported File Formats

Input Images

• JPEG (.jpg, .jpeg)
• PNG (.png)
• BMP (.bmp)
• TIFF (.tiff, .tif)

Output Files

• Masks: PNG binary images (black/white)
• Overlays: PNG/JPEG with colored mask overlay
• Logs: UTF-8 text files

⚡ Keyboard Shortcuts

• Ctrl+O: Load image (when available)
• Ctrl+S: Save current mask (when available)
• Escape: Clear current annotations
• Space: Generate mask (when inputs are ready)

🐛 Troubleshooting

Common Issues

"请先加载模型" (Please load model first)

• Ensure you've selected a model folder with valid .pth files
• Check that the model architecture matches your files

"SAM模块不可用" (SAM modules unavailable)

• Install required dependencies: pip install torch opencv-python pillow
• Ensure SAM model files are in the correct directory

Slow performance

• Try ViT-B model for faster processing
• Reduce image resolution if very large
• Close other applications to free memory

Auto segmentation fails

• Check if your model supports automatic mask generation
• Try Point or Box mode instead
• Ensure sufficient system memory

Getting Help

1. Check the log panel for detailed error messages
2. Verify model files are correctly named and not corrupted
3. Try different operation modes if one isn't working
4. Restart the application if interface becomes unresponsive

---

📋 Quick Reference

操作	步骤
基础分割	加载模型 → 加载图像 → 点击模式 → 点击对象 → 生成mask
快速框选	加载模型 → 加载图像 → 框选模式 → 拖拽矩形框
自动发现	加载模型 → 加载图像 → 自动分割 → 生成mask
保存结果	生成mask → 保存mask（二进制）或 保存叠加图（可视化）
重新开始	点击清除 → 添加新的点/框 → 生成mask

💡 专业提示: 日志信息面板会显示后台具体操作过程 - 如果遇到问题请检查日志信息！