joonhaim/hinged-state-classifier-crops

---
license: mit
task_categories:
- image-classification
language:
- en
tags:
- computer-vision
- image-classification
- surgical-tools
- surgical-instruments
- medical-imaging
- open-closed-classification
pretty_name: hinged-state-classifier-crops
size_categories:
- 1K<n<10K
---

# hinged-state-classifier-crops

Binary image classification dataset for **open vs. closed** state recognition of **hinged surgical instruments**.

## Overview

This dataset contains cropped images of three hinged surgical tool categories:

- **CM** = clamp
- **NH** = needle holder
- **SH** = shear

The target classes are:

- **open**
- **closed**

It is intended for training and evaluating hinge-state classifiers in a two-stage surgical tool recognition pipeline, where an instrument is first detected and then classified as open or closed.

## Structure

hinged_state_classifier_crops/
├── train/
│ ├── open/
│ └── closed/
├── val/
│ ├── open/
│ └── closed/
└── test/
├── open/
└── closed/

## Dataset size

Total images: **1512**

| Split | Open | Closed | Total |
|---|---:|---:|---:|
| Train | 603 | 606 | 1209 |
| Val | 75 | 71 | 146 |
| Test | 78 | 79 | 157 |
| **Total** | **756** | **756** | **1512** |

The dataset is fully balanced at the class level.

## Global distribution

### By tool

| Tool | Count |
|---|---:|
| CM | 504 |
| NH | 504 |
| SH | 504 |

### By view

| View | Count |
|---|---:|
| CLO | 504 |
| OBL | 504 |
| TOP | 504 |

### By background

| Background | Count |
|---|---:|
| BL | 378 |
| GR | 378 |
| TR | 378 |
| WH | 378 |

## Filename convention

Images follow this naming pattern:

{TOOL}_{VIEW}_{BACKGROUND}_{STATE}_{INDEX}

Where:

- `TOOL`: `CM`, `NH`, `SH`
- `VIEW`: `CLO`, `OBL`, `TOP`
- `BACKGROUND`: `BL`, `GR`, `TR`, `WH`
- `STATE`: `OP`, `CL`

## Validation

The dataset passed validation successfully:

- Expected total images: 1512
- Observed total images: 1512
- Problems found: 0
- Warnings found: 0
- Duplicate image-content groups: 0
- Cross-split duplicate image-content groups: 0

## Intended use

This dataset is intended for:

- binary image classification of hinged surgical instrument state,
- benchmarking open/closed classifiers,
- evaluating robustness across viewpoints and backgrounds,
- use in multi-stage surgical tool recognition systems.