DWposeDeluxe Estimator (No TRT)

<div id="dwpdlx_short_desc">Main node for DWpose, detects and renders human poses.</div> **Function:** The main node that detects and renders human poses from an input image sequence. ### Inputs - **image:** The source image frames to process. - **frame_count:** The total number of frames in the sequence. - **audio:** Optional audio to pass through with the video. - **video_info:** Optional video metadata for frame rate handling. - **custom_options:** Optional advanced rendering and detection parameters. ### Parameters - **poses_to_detect:** Max number of poses to detect per frame (by bbox area). - **show_body:** Toggle visibility of the main body skeleton. - **show_feet:** Toggle visibility of the foot points and bones. - **show_face:** Toggle visibility of face keypoints. - **show_hands:** Toggle visibility of hand points and bones. - **provider_type:** Hardware backend for estimation (CPU or GPU/TensorRT). - **precision:** Floating point precision for GPU models. - **detector_model:** The person detector model to use. - **estimator_model:** The pose estimator model to use. - **save_keypoints:** Save the output keypoint data to a file. ### Outputs - **pose_image:** The rendered pose skeletons on a black background. - **blend_image:** The pose skeletons blended over the source image. - **face_image:** A horizontal atlas of cropped faces (left to right). - **source_image:** The original input image, passed through for easy noodling. - **audio:** The original input audio, passed through from video (up)load node. - **frame_rate:** The original video frame rate, passed through from video_info. - **keypoints:** The raw detected keypoint data in JSON format. ### Notes - Max poses_to_detect is far from perfect and can be glitchy on high action videos with multiple foreground poses but works like charm to exclude unwanted background photo-bombers. A more robust function coming soon! - The face_crop images are concatenated in left to right order, so will re-arrange if characters swap places during video sequence. Fix coming soon! Workaround: range_from_batch, crop and re-batch nodes are your friends.