Conditioning maps
From the photo camera, one size and framing, without lens distortion: an OpenPose skeleton (COCO-18, with its keypoints as JSON), depth (white near; 16-bit millimetres in the zip), camera-space normals, line art, the subject's silhouette and a mask per object. The zip's manifest.json says how each is encoded, with the near and far depth and the camera's intrinsics. Click a map to download it.