vision

vision copied to clipboard

Added the detection and binary-detection target types to the Oxford III pet data set loader as discussed in #8364

The dataset loader will load bounding box annotations in the format expected by torchvision's faster RCNN network. It will also filter out images that exist in the dataset that do not include bounding box annotations when one of the detection types is selected. A message is printed to the console indicating that some images are filtered in this case.

Demonstration of the functionality of the detection target types can be found here and here.

Looking for feedback on 2 things:

1. Given that some images have to be filtered out for this to work, is a simple print statement the best way to notify users?
2. Given that the _to_rcnn function should work with any dataset using VOC-style detection labels, should we move it to the VOC loader and import it instead of defining it here?

:link: Helpful Links

:test_tube: See artifacts and rendered test results at hud.pytorch.org/pr/pytorch/vision/8425

• :page_facing_up: Preview Python docs built from this PR

Note: Links to docs will display an error until the docs builds have been completed.

This comment was automatically generated by Dr. CI and updates every 15 minutes.