-
Notifications
You must be signed in to change notification settings - Fork 0
Open
Description
Task:
- Build a small tool to extend the current dataset with handle orientation annotations.
- Current label format (YOLO default, from Roboflow):
<class_id> <x_center> <y_center> <width> <height>Example:
0 0.512 0.478 0.220 0.330- Proposed expanded format (with orientation θ):
<class_id> <x_center> <y_center> <width> <height> <theta>Example:
0 0.512 0.478 0.220 0.330 1.5708- Required functionality::
- Load image with its bounding box (Roboflow export).
- User draws a line along the valve handle.
- Tool calculates orientation angle θ (in radians, CCW).
- Save angle alongside bbox in the updated label file.
- Keep it lightweight (Python + OpenCV or Matplotlib GUI is fine).
Angle convention (must be consistent)
- 0° / 0 rad -> Valve handle pointing upwards (Shut position).
- 90° / π/2 rad -> Valve handle pointing left (Open position).
- Rotation is counterclockwise (CCW) from Shut -> Open.
For example, in the image below the handle is vertical: θ = 0 rad (0°)

Deliverables:
- Simple GUI tool or script for manual angle annotation.
- Updated dataset format:
bbox + θ. - Documentation:
- How to use the tool.
- Example of annotated label files before/after.
References:
- OpenCV drawing functions
- Matplotlib interactive events
- Roboflow custom annotation docs
- Roboflow download example (see Step 1 in
train.pyfor how to fetch images + YOLO Labels from Roboflow)
Reactions are currently unavailable
Metadata
Metadata
Assignees
Labels
No labels