Great stuff! I miss the node for HandsDetector…
indeed, VL.MediaPipe 0.1.0-alpha is now up with parsers for all models! talked a bit about it at last nights meetup.
6 Likes
is it possible to use the live image from a webcam as a texture while tracking with MediaPipe?
you’d have to use SpoutCam as described here: GitHub - torinmb/mediapipe-touchdesigner: GPU Accelerated MediaPipe Plugin for TouchDesigner
1 Like
Hi, tanks for the nuget, it would be possible to select a single element from the various detectors as seen in the touchdesigner video? (Like the position of a specific finger or similar)
Yeah, just dig for the finger or joint you need:
The second example in the nuget paints squares on all fingers, but you can of course filter for the one you want. Each hand has 21 including the wrist.
I did this for Genuary yesterday and it is really cool:
What I found is, that one needs to have a bright room for it to be fast. Low light needed much more detection time while a bright scene was up to 5 times faster.
Also a cleaner image seems to detect faster, my GFs webcam is very noisy and that was slower and more jittery.
I drew points between the joints for visualisation first, then used them as emitters:
1 Like
VL.MediaPipe 0.1.2-alpha now comes with PoseJointByName and FingerJointByName nodes allow you to select the respective joints via enum istead of index.
4 Likes
Great, thanks!
VL.MediaPipe 0.3.0-alpha:
- breaking change: simplified working with Hands, FingerJoints providing nodes: SelectHand, SelectJoint
- updated to underlying mediapipe-touchdesigner 0.4.0 which adds new ImageEmbedder feature
3 Likes
This is super cool!
Ive had a look ad the Finger tracking data and its 2d only, it looks like you need to sample a depthmap to insert the Z-value for each joint. ( mediapipe/docs/solutions/hands.md at master · google/mediapipe · GitHub )
Is it possible to output the depthmap?
Edit: Never mind got it working more more or less, will post a patch
1 Like
The HandDetector returns Landmarks and WorldLandmarks the latter of which has non-zero values for Z. is that also what you found?
1 Like
yes not sure if im doing it right but it looks like you have to combine them. Landmarks + Worldlandmarks with aspect ratio correction. its matching but not perfect
2 Likes
Hello! I’m working on a project where I need to implement face recognition of a person’s name. I’m considering using Mediapipe for this task. Can anyone share their experience or advice on using Mediapipe for recognizing people’s names? I would be grateful for any guidance or code examples. Thank you!