Datasets
The PASCAL and COCO datasets that were mentioned in Chapter 4, Object Detection, can be used for the segmentation task as well. The annotations are different as they are labelled pixel-wise. New algorithms are usually benchmarked against the COCO dataset. COCO also has stuff datasets such as grass, wall, and sky. The pixel accuracy property can be used as a metric for evaluating algorithms.
Note
Apart from those mentioned, there are several other datasets in the areas of medical imaging and satellite imagery. The links to a few of them are provided here for your reference:
- http://www.cs.bu.edu/~betke/BiomedicalImageSegmentation
- https://www.kaggle.com/c/intel-mobileodt-cervical-cancer-screening/data
- https://www.kaggle.com/c/diabetic-retinopathy-detection
- https://grand-challenge.org/all_challenges
- http://www.via.cornell.edu/databases
- https://www.kaggle.com/c/dstl-satellite-imagery-feature-detection
- https://aws.amazon.com/public-datasets/spacenet
- https://www.iarpa.gov/challenges/fmow.html
- https...
