Examples

Below are examples of predictions with OpenPifPaf’s command line tool openpifpaf.predict. These examples can serve as a starting point to create pose predictions on your own images.

COCO

Below are example predictions from the COCO val set. These are the first images with more than five person annotations that were shared with CC-BY-2.0. These examples are not cherry-picked.

All examples are predicted with the fast model shufflenetv2k16. For more accurate results, try --checkpoint=shufflenetv2k30.

%%bash
python -m openpifpaf.predict coco/000000188465.jpg -q --image-output --long-edge=1025

import IPython
IPython.display.Image('coco/000000188465.jpg.predictions.jpeg')

Image credit: Photo by Brian Jelonek shared under CC-BY-2.0.

%%bash
python -m openpifpaf.predict coco/000000114907.jpg -q --image-output

IPython.display.Image('coco/000000114907.jpg.predictions.jpeg')

Image credit: Photo by Fort Rucker shared under CC-BY-2.0.

%%bash
python -m openpifpaf.predict coco/000000147740.jpg -q --image-output

IPython.display.Image('coco/000000147740.jpg.predictions.jpeg')

Image credit: Photo by CCFoodTravel.com shared under CC-BY-2.0.

%%bash
python -m openpifpaf.predict coco/000000541055.jpg -q --image-output

IPython.display.Image('coco/000000541055.jpg.predictions.jpeg')

Image credit: Photo by Jarrod Doll shared under CC-BY-2.0.

Debug Outputs

%%bash
python -m openpifpaf.predict coco/000000541055_cropped.jpeg -q --image-output --debug-indices cif:5 cifhr:5 caf:5 --save-all=coco_debug_example

IPython.display.Image('coco_debug_example/0001.jpeg')  # cif:5:confidence

IPython.display.Image('coco_debug_example/0002.jpeg')  # cif:5:regression

IPython.display.Image('coco_debug_example/0005.jpeg')  # cifhr:5

IPython.display.Image('coco_debug_example/0003.jpeg')  # caf:5:confidence

IPython.display.Image('coco_debug_example/0004.jpeg')  # caf:5:regression

Image credit: Photo by Jarrod Doll shared under CC-BY-2.0.