Buckets:

---

Atharv1103

/

PartImageNetPP-bucket

PartImageNet++ Dataset

PartImageNet++ is an extensive dataset designed for robust object recognition and segmentation tasks. This dataset expands upon the original ImageNet dataset by providing detailed part annotations for each object category. Official repository for PartImageNet++ Dataset: Scaling up Part-based Models for Robust Recognition.

Dataset Statistics

The dataset includes:

• 1000 object categories derived from the original ImageNet-1K.
• 3308 part categories representing different parts of objects.
• 100,000 annotated images, with each object category containing 100 images (downloaded from the ImageNet-1K dataset).
• 406,364 part mask annotations ensuring comprehensive coverage and detailed segmentation.

Structure and Contents

Each JSON file in the json directory represents one object category and its corresponding part annotations.

The including folder provides detailed inclusion relations of parts, illustrating hierarchical relationships between different part categories.

The discarded_data.json file lists low-quality images excluded from the dataset to maintain high annotation standards.

The category_name.json file contains each JSON file's file name, along with its corresponding part name and object name.

Visualizations

We provide a visualization demo tool to explore and inspect the annotations. This tool helps users better understand the dataset's structure and details.

If you find this useful in your research, please cite this work:

@inproceedings{li2024pinpp,
  author = {Li, Xiao and Liu, Yining and Dong, Na and Qin, Sitian and Hu, Xiaolin},
  title = {PartImageNet++ Dataset: Scaling up Part-based Models for Robust Recognition},
  booktitle={European conference on computer vision},
  year = {2024},
  organization={Springer}
}