Surgical Instrument Images for Multiview Object Detection

2520 images of individual surgical instruments across five tool classes: clamp, needle_holder, scalpel, shear, and tweezer. The dataset, created by joonhaim, is split into 2016 training, 252 validation, and 252 test images. It was last updated in April 2026.

Use Cases

• Train a YOLO-based object detector to predict bounding boxes and class labels (e.g., 'scalpel', 'tweezer') from surgical images.
• Benchmark multiview recognition models using the provided image splits (train, validation, test) for each tool class.
• Develop data augmentation strategies specific to surgical instruments by leveraging the five distinct visual classes.
• Fine-tune a pre-trained convolutional neural network on the 2016 training images to classify surgical tools.

Strengths

• 2520 total images provide a foundation for model training.
• Five distinct tool classes (clamp, needle_holder, scalpel, shear, tweezer) enable multi-class recognition tasks.
• Pre-defined splits of 2016 training, 252 validation, and 252 test images support standardized evaluation.

Limitations

• Sample size of 2520 images is modest for deep learning, potentially limiting model generalization.
• Limited to five tool classes, which may not represent the full diversity of surgical instruments.
• Dataset size and geographic or institutional origin of the images are not specified, raising questions about bias.

Provenance

Source

Author joonhaim on Hugging Face.

Collection Method

Originally created in YOLO format for object detection; contains images of individual surgical instruments.

Freshness

Last updated in April 2026.