Following up on https://github.com/pytorch/vision/pull/5876#discussion_r857726932 The "Detection" section of the new models doc contains `KeypointRCNN`, but it should be treated separately as it has an additional keypoint_MAP metric.