TouchDesigner

TrackOSC receiver example for TouchDesigner

TouchDesigner's .toe files are binary, so this example ships the parts that can live in a text repository – a Python callbacks script and a recipe for the network – rather than a project file. Building it takes about five minutes.

File	Paste into
trackosc_callbacks.py	The Callbacks DAT of an OSC In DAT. Parses every TrackOSC message into Table DATs.
trackosc_skeletons.py	A Text DAT named trackosc_skeletons (joint names and edge lists, imported by the callbacks).
trackosc_script_sop.py	Optional: the callbacks DAT of a Script SOP, to draw skeletons as geometry.

Why an OSC In DAT rather than an OSC In CHOP

TrackOSC messages have a variable number of arguments (n detections, and /faces/contour has a per-face point count) and carry strings (text, animal labels, barcode payloads). An OSC In CHOP flattens each message into fixed channels and drops strings, which works only for /camerainfo. The OSC In DAT hands the raw argument list to Python, where the same running-cursor parser as every other example turns it into tables – and tables convert to CHOPs, SOPs or anything else.

Recipe

1. OSC In DAT (oscin1): Network Port 9527, Address Scope *, Bundle Timestamp off. Only one program can listen on a port – quit the native TrackOSC Receiver first.
2. Text DAT named exactly trackosc_skeletons: paste trackosc_skeletons.py into it.
3. Open oscin1's Callbacks DAT (the oscin1_callbacks Text DAT it creates) and replace its contents with trackosc_callbacks.py.
4. Table DATs, one per message you want, named exactly: camerainfo, poses, hands, faces, faces_box, faces_contour, texts, animals, poses3d, barcodes, animalposes, humans, and frames. Tables that don't exist are skipped, so start with poses and frames. Each fills with a header row and one row per joint / point / detection; frames holds the latest width, height and n per address.
5. Point a TrackOSC sender at this machine's IP, port 9527 (or run swift run poseiosc-testsend 127.0.0.1 9527 from the repository, or python3 Examples/Python/trackosc_testsend.py without Xcode). The tables update live.
6. To get channels: DAT to CHOP on poses with the x, y, c columns selected, one sample per row (row = joint, see the name column).
7. To draw: a Script SOP with trackosc_script_sop.py as its callbacks, inside a Geometry COMP with a Line MAT, gives the body skeleton as line geometry in normalised coordinates; a Camera COMP + Render TOP shows it. Change TABLE/COUNT in the script for hands or animals.

Table layouts

Table	Columns
poses, hands, faces, animalposes	det joint name x y c conf – pixels in the sent frame, origin top-left; c == 0 means missing
texts, animals	det conf left top width height label
humans	det conf left top width height
faces_box	det conf left top width height roll yaw pitch (degrees)
faces_contour	det point x y conf – an open jawline polyline; a face with no contour has no rows
poses3d	det joint name x y z px py conf bodyHeight – x/y/z metres (Vision camera space), px/py pixels
barcodes	det conf left top width height tl_x tl_y tr_x tr_y br_x br_y bl_x bl_y symbology payload
camerainfo	key value rows: width, height, orientation (0/90/180/270), facing (0 back, 1 front), time
frames	address width height n time

Joint orders and edge lists: Examples/SKELETONS.md. Wire format: the root README.

Verification

The parser is exercised outside TouchDesigner by python3 -m unittest discover tests (stubbing op() and mod). The network recipe was written from TouchDesigner's documentation, not run in TouchDesigner by the author – if you build it, please open an issue or pull request with the TouchDesigner version you used (and a .toe would be very welcome).