Data labeling specialist to create a training dataset
Budget: $10 – $30 CAD
I'm looking for a data labeling specialist to create a training dataset for detecting electrical components in residential building plans. The plans are in PDF format (exported from AutoCAD) and can also be converted to images (PNG or JPG) for processing and labeling. The ultimate goal is to train a YOLOv8 or similar model to automatically recognize electrical symbols. # Specific tasks: - Convert PDF or DWG plans to images - Extract each plan (one per image) - Maintain optimal resolution and readability - Data labeling (manual or semi-automatic)
- Use Roboflow, CVAT, or LabelImg - Create object classes (minimum 8–10 electrical types) - Draw precise bounding boxes over each symbol - Export the final dataset - Format: YOLOv8 or COCO JSON - Divided into train/val/test - Clean naming and structure (e.g., images/train/, labels/train/) - Implement an initial detection test - Train the YOLOv8 model with 100–200 examples - Verify basic accuracy. # Estimated amount of data required: Unique electrical drawings (PDF) 100 to 150 drawings; Resulting images (one per drawing or view) 300–500 images; Total expected labels 15,000–25,000 objects (across all drawings); Classes (symbol types)
8–12 classes (e.g., outlet, switch, light, fan, panel, data port...). Each symbol must be labeled manually or semi-automatically with precise bounding boxes. # Expected result: A professional dataset for training YOLOv8 models, enabling the automatic detection of electrical components in construction drawings.
- Use Roboflow, CVAT, or LabelImg - Create object classes (minimum 8–10 electrical types) - Draw precise bounding boxes over each symbol - Export the final dataset - Format: YOLOv8 or COCO JSON - Divided into train/val/test - Clean naming and structure (e.g., images/train/, labels/train/) - Implement an initial detection test - Train the YOLOv8 model with 100–200 examples - Verify basic accuracy. # Estimated amount of data required: Unique electrical drawings (PDF) 100 to 150 drawings; Resulting images (one per drawing or view) 300–500 images; Total expected labels 15,000–25,000 objects (across all drawings); Classes (symbol types)
8–12 classes (e.g., outlet, switch, light, fan, panel, data port...). Each symbol must be labeled manually or semi-automatically with precise bounding boxes. # Expected result: A professional dataset for training YOLOv8 models, enabling the automatic detection of electrical components in construction drawings.