OpenDetection-80K-Unified-Cleaned

## **OpenDetection-80K-Unified-Cleaned** **OpenDetection-80K-Unified-Cleaned** is a large-scale object detection dataset built primarily from general, publicly available images, which make up the majority of the input imagery, together with additional publicly available datasets. This dataset is a unified collection created by combining **OpenDetection-15K-Dense-v1.0**, **OpenDetection-15K-Dense-v2.0**, and **OpenDetection-50K-Remastered-Cleaned** into a single standardized dataset. Every sample contains at least one valid object annotation, ensuring there are no empty detection records. The dataset provides structured object detection annotations consisting of class labels, label IDs, confidence scores, and bounding box coordinates, along with an annotated visualization of every image. It is intended for training, evaluation, benchmarking, and multimodal computer vision research, while its optimized Parquet format enables efficient loading for large-scale machine learning workflows. ## Dataset Statistics | Property | Value | |-----------|-------| | Number of Samples | 82,379 | | Image Format | RGB | | Annotation Format | JSON | | Visualization | Annotated Image | | Dataset Format | Optimized Parquet | ## Dataset Composition The dataset is created by unifying the following datasets: | Dataset | Description | |---------|-------------| | [OpenDetection-15K-Dense-v1.0](https://huggingface.co/datasets/prithivMLmods/OpenDetection-15K-Dense-v1.0) | Object detection dataset with dense annotations | | [OpenDetection-15K-Dense-v2.0](https://huggingface.co/datasets/prithivMLmods/OpenDetection-15K-Dense-v2.0) | Second curated dense object detection dataset | | [OpenDetection-50K-Remastered-Cleaned](https://huggingface.co/datasets/prithivMLmods/OpenDetection-50K-Remastered-Cleaned) | Cleaned object detection dataset with all empty annotations removed | The resulting unified dataset contains consistent annotations and preserves the same data structure across all samples. ## Dataset Structure Each sample contains the following fields: | Column | Type | Description | |---------|------|-------------| | `image` | Image | Original input image | | `objects` | List | Object detection annotations containing labels, label IDs, confidence scores, and bounding boxes | | `annotated_image` | Image | Visualization of the image with rendered bounding boxes | Example: ```python sample = ds[0] print(sample.keys()) # dict_keys([ # "image", # "objects", # "annotated_image" # ]) ``` ## Loading the Dataset ```python from datasets import load_dataset ds = load_dataset( "prithivMLmods/OpenDetection-80K-Unified-Cleaned", split="train" ) ``` ## Example Usage ```python from datasets import load_dataset import matplotlib.pyplot as plt ds = load_dataset( "prithivMLmods/OpenDetection-80K-Unified-Cleaned", split="train" ) sample = ds[0] image = sample["image"] objects = sample["objects"] annotated = sample["annotated_image"] print("Detected Objects:") print(objects) fig, axes = plt.subplots(1, 2, figsize=(12, 6)) axes[0].imshow(image) axes[0].set_title("Image") axes[0].axis("off") axes[1].imshow(annotated) axes[1].set_title("Annotated Image") axes[1].axis("off") plt.show() ``` ## Dataset Features - Unified collection of three OpenDetection datasets - More than 82K annotated images - Every image contains at least one valid object annotation - High-quality structured object detection annotations - Bounding box coordinates and confidence scores for every detected object - Annotated visualization for every image - Optimized Parquet format for efficient loading - Compatible with the Hugging Face Datasets library - Suitable for object detection training, evaluation, benchmarking, and multimodal computer vision research ## Citation If you use this dataset in your research or projects, please cite: ```bibtex @dataset{OpenDetection80KUnifiedCleaned2026, title={OpenDetection-80K-Unified-Cleaned}, author={Prithiv Sakthi}, year={2026}, publisher={Hugging Face}, url={https://huggingface.co/prithivMLmods} } ``` Dataset maintained by **Prithiv Sakthi ([prithivMLmods](https://huggingface.co/prithivMLmods))**. ## License This dataset is released under the **Apache-2.0 License**.