Engine Assembly Images with 8 Object Classes, 280k Annotated Frames

195 real images of engine assemblies were captured and annotated with bounding boxes, keypoints, and polygons. The dataset was expanded to approximately 280,000 images through augmentation. Gaurang Sharma from Tampere University created this dataset, which includes generation tools and a trained model.

Use Cases

• Train object detection models based on the 8 annotated object classes.
• Develop keypoint estimation models for engine components based on the provided annotations.
• Generate synthetic training data for manufacturing inspection using the provided augmentation scripts.
• Benchmark multimodal (bounding box, keypoint, polygon) annotation performance on industrial imagery.

Strengths

• Includes three annotation types: bounding boxes, keypoints, and polygons.
• Scale increased from 195 real images to approximately 280,000 images via augmentation.
• Provides all tools and scripts for dataset generation and replication.
• Includes a trained model and visualization scripts for immediate use.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Last update date is unknown; freshness unverified.
• The dataset's origin from 195 real images may reflect specific environmental or assembly biases.

Provenance

Source

Tampere University

Collection Method

Real images were captured and annotated, then augmented.