Image classification, object detection, image segmentation and instance segmentation datasets containing 37 categories of cats and dogs.
Original dataset: images (755MB), annotations (18MB)
Image classification dataset (752MB)
Image segmentation dataset (animal ROI; bluechannel) (769MB)
Object detection dataset (head bbox; VOC XML format) (359MB)
Object detection dataset (animal bbox; VOC XML format) (710MB)
Object detection/instance segmentation dataset (animal ROI; ADAMS) (726MB)
Conversion from original data:
Download the images and annotations archives and extract them, resulting in the annotations and images directories.
Create output directory at the same level as extracted archives.
-
Create virtual environment using these requirements:
python3 -m venv venv
./venv/bin/pip install -r requirements.txt
-
Classification
Run the classification.py Python script: ./venv/bin/python classification.py . output
The output/subdir directory contains the image classification dataset
-
Object detection (head ROI)
Run the object_detection.py Python script: ./venv/bin/python object_detection.py . output
The output/voc-head directory contains the object detection dataset with the head ROI annotations
-
Object detection, image/instance segmentation (animal ROI)
Run the segmentation.py Python script: ./venv/bin/python segmentation.py . output
The output/bluechannel directory contains the image segmentation dataset
The output/adams directory contains the instance segmentation dataset in ADAMS format
-
Convert the ADAMS annotations into MS COCO using the wai.annotations library:
wai-annotations convert from-adams-od -i "./adams/*.report" to-coco-od -o ./coco/annotations.json --sort-categories --category-output-file labels.txt -
Convert the ADAMS annotations into VOC using the wai.annotations library:
wai-annotations convert from-adams-od -i "./adams/*.report" coerce-box to-voc-od -o ./voc-animal
License
Citation