Object Detection Sample
For object detection samples, different categories of target objects need to be marked in sample images with corresponding annotations. SuperMap iDesktopX provides target object type management and labeling functions, along with four annotation drawing methods: rectangle, polygon, line buffer polygon, and batch drawing.
Steps
- Create/Open Object Detection Sample Library: Prepare sample library and object types. For library creation, opening or import methods, refer to Sample Library Management.
- Draw Sample Tags:
- AI Label: Use PaddleSeg or SegmentAnything models to quickly extract samples. Details see AI Label.
- Rectangle: Select Rectangle from the Polygon dropdown menu. Move cursor to sample image window to draw rectangular labels.
- Polygon: For precise object boundaries, choose Polygon from dropdown menu. Draw polygonal labels in sample image window.
- Line: For linear-distributed targets, draw polylines and generate buffer polygons with specified radius for labeling.
- Batch Draw: For similarly-sized objects, batch add fixed-size rectangular labels by setting width, height and type, then click target positions.
- After drawing tags, the sample library list displays label counts per image.
- Edit Sample Tags:
- Delete: Select tag in image window or label list, then use Delete key or button to remove.
- Adjust Tag Size: Select tag nodes to modify positions. Left-click on edges to insert new nodes.
- Modify Tag Type: Select tag in label extraction window, click Edit button, then choose new type in dialog.
- Undo: Use Ctrl+Z to reverse add/delete/edit operations.
- Sample Label Settings
- Buffer Radius: Set buffer radius (in pixels) for line-to-polygon conversion.
- Display Label Classification: Toggle class name display in image window.
- Layer Style: Click Layer Style to configure fill color and border style in Fill Symbol Selector.
- Difficult to Identify: Mark selected tags as difficult samples that participate in training but not accuracy evaluation.
Related Topics