Skip to content

AnnoMate.md

Latest commit

 

History

History
65 lines (53 loc) · 4.44 KB

AnnoMate.md

File metadata and controls

65 lines (53 loc) · 4.44 KB

AnnoMate User Guide

AnnoMate is the core manual annotation and review component of the suite. It allows users to create ground-truth segmentations, review part quality, manage project files, and export data.

1. Interface Overview

  • Left Canvas: Your primary workspace. Here you can zoom (Scroll Wheel), pan (Right-Click + Drag), and draw polygons.
  • Left Tool Palette: Contains the Polygon tool, SAM 2 tool, and Brush Thickness controls.
  • Right Panel: Contains the Dataset Navigator, Annotation Classes, Current Image Annotations, and Metadata sections.

2. Project Management (.annoproj)

AnnoMate uses a robust project system. A .annoproj file saves your images, class definitions, annotations, inspector notes, and the path to your loaded AI model all in one place.

Creating a New Workspace

  1. Go to File > Open Image Folder... and select a local directory containing your dataset (.jpg, .png, .bmp, .tif).
  2. The images will load into the Dataset Navigator on the right panel.

Saving Your Progress

  1. Go to File > Save Project As...
  2. Choose a location and name for your project. This will generate your .annoproj file and an associated annotations.coco.json file.
  3. Autosave: Once a project is saved, the application will automatically create an autosave backup inside your project folder every 5 minutes while you work.

Relocating Images

Annotations are saved using relative paths to your image folder. If you ever move your image folder on your hard drive, your .annoproj file will warn you that the images are missing.

  • Fix this by going to File > Relocate Images... and selecting the new folder location. Your annotations will immediately map to the new file paths.

3. The Dataset Navigator

The top of the right panel displays your loaded images.

  • Navigation: Click a row or use the Prev/Next buttons to switch images.
  • Status Dots:
    • 🟢 Green (Reviewed): The image has annotations, an inspector name, a note, or a final Accept/Reject decision.
    • 🟠 Orange (Pending): The image has not been interacted with.

4. Creating Annotations

⬠ Polygon Tool (Shortcut: P)

Used for manual, point-by-point drawing.

  1. Select a class from the Annotation Classes list.
  2. Click the Polygon Tool (⬠).
  3. Left Click: Place vertices on the canvas.
  4. Backspace: Undo the last vertex placed.
  5. Double-Click (or click the start point): Close and save the polygon.
  6. Escape: Cancel the current drawing.

✦ SAM 2 Bounding Box Tool (Shortcut: S)

Uses Meta's Segment Anything 2 AI to automatically generate precise polygon masks. (Note: Requires internet access on first use to download model weights).

  1. Click the SAM Tool (✦).
  2. Left Click & Drag: Draw a bounding box tightly around the defect.
  3. The AI generates a dashed "Ghost Polygon" preview.
  4. Enter: Accept the ghost polygon as an annotation.
  5. Escape: Reject the ghost polygon.
  • Tip: Click the gear icon (⚙) below the SAM tool to switch variants. "Tiny" is fastest; "Large" is most accurate.

5. Editing & Reviewing

  • Modify Shapes: With no tool selected, click inside a polygon to drag the entire shape, or click and drag a specific vertex (dot) to adjust its outline.
  • Adjust Thickness: Click the Brush Thickness (◢) icon. If a polygon is selected, the slider adjusts that specific polygon's line thickness.
  • Delete: Select a polygon and press the Delete key, or click the Trash icon in the "Current Image Annotations" list.
  • Accept/Reject Part: Use the floating ✓ Accept or ✗ Reject buttons at the top right of the canvas to mark the part's quality.
  • Inspector Notes: Use the bottom-right Metadata section to log your name and any notes about the part.

6. Exporting & Importing Data

Found under the Data menu:

  • Export Polygons + Data: Saves a _data.json file containing all polygon coordinates and metadata. It also generates JPEG images with the annotations permanently "burned in" for sharing.
  • Export Binary Masks: Renders pure black-and-white .png images (defects are white, background is black). This is the standard format required for training AI models.
  • Export CSV: Generates a spreadsheet containing the Image Name, Inspector, Notes, Accept/Reject decision, and classes present.
  • Import JSON Data: Loads previously exported AnnoMate JSON files or standard COCO JSON files directly into your workspace.