pycocoedit is a Python package for editing and analyzing COCO datasets.
It is particularly useful for specifying which images, annotations, or categories to include or exclude from your dataset.
With pycocoedit, you can apply custom filters to your dataset. These filters allow you to control inclusion or exclusion based on custom conditions for images, categories, and annotations.
For example, you can filter out specific images that have a certain number of annotations or exclude annotations with bounding boxes of a certain aspect ratio.
Example of filtering images and categories.
from pycocoedit.objectdetection.data import CocoData
from pycocoedit.objectdetection.filter import FilterType, ImageFileNameFilter, CategoryNameFilter
annotation = "path/to/annotation.json"
new_annotation = "path/to/new_annotation.json"
# only include images with file name "image1.jpg" and "image2.jpg"
file_filter = ImageFileNameFilter(FilterType.INCLUSION, ["image1.jpg", "image2.jpg"])
# only include categories with category name "cat" and "dog"
category_filter = CategoryNameFilter(FilterType.INCLUSION, ["cat", "dog"])
coco_data = CocoData(annotation)
# apply filters and export new annotation
coco_data.add_filter(file_filter).add_filter(category_filter).apply_filter().save(new_annotation)Example of custom filter for annotations:
In this example, we create a custom filter that only includes annotations with bounding boxes of area less than 100.
from pycocoedit.objectdetection.data import CocoData
from pycocoedit.objectdetection.filter import BaseFilter, FilterType, TargetType
# only include annotations with area less than 100
class SmallBboxIncludeFil
7708
ter(BaseFilter):
def __init__(self):
super().__init__(FilterType.INCLUSION, TargetType.ANNOTATION)
def apply(self, data: dict) -> bool:
return data["area"] < 100
annotation = "path/to/annotation.json"
new_annotation = "path/to/new_annotation.json"
coco_data = CocoData(annotation)
# apply custom filter and export new annotation
coco_data.add_filter(SmallBboxIncludeFilter()).apply_filter().save(new_annotation)pip install pycocoedit
| Feature | What it gives you |
|---|---|
| LEGO-style chainable filters | One-liner include / exclude rules for images, annotations, categories, etc. |
| Custom rules | simply inherit BaseFilter, implement a short apply() method, and your custom logic is ready to use. |
| Built-in data cleanup | CocoData.correct() - Built-in data cleanup that removes orphaned annotations & empty categories. |
| Typed & unit-tested | IDE auto-completion and high confidence when refactoring. |
| Task | Supported | version |
|---|---|---|
| Object Detection | ✅ (pycocoedit.objectdetection) |
0.0.1 |
| Image Segmentation | ✅ (use pycocoedit.objectdetection) |
0.0.1 |
| Keypoint Detection | ❌ (future release) | |
| Panoptic Segmentation | ❌ (future release) | |
| Image Captioning | ❌ (future release) |
- Image Captioning
- key-point support
- Panoptic Segmentation
Contributions and ideas are welcome—feel free to open an issue or pull request on GitHub!
This project is licensed under the Apache License 2.0 - see the LICENSE file for details.