google-ai-edge / mediapipe

Cross-platform, customizable ML solutions for live and streaming media.
https://ai.google.dev/edge/mediapipe
Apache License 2.0
27.39k stars 5.15k forks source link

Hand World Landmarks are not accurate #5571

Open Javacr opened 2 months ago

Javacr commented 2 months ago

Have I written custom code (as opposed to using a stock example script provided in MediaPipe)

None

OS Platform and Distribution

windows 10

MediaPipe Tasks SDK version

0.10.10

Task name (e.g. Image classification, Gesture recognition etc.)

Gesture recognition

Programming Language and version (e.g. C++, Python, Java)

python

Describe the actual behavior

Hand 2D landmarks are accurate, but hand world landmarks are bad

Describe the expected behaviour

Hand world landmarks are accurate

Standalone code/steps you may have used to try to get what you need

I follow https://ai.google.dev/edge/mediapipe/solutions/vision/gesture_recognizer/python to detect hand gesture, and I plot 2D and 3D hand landmarks. The 3D hand landmarks are weird.

Other info / Complete Logs

No response

Javacr commented 2 months ago

image image image image red represents left hand, blue represents right hand

Javacr commented 2 months ago

image image image image

In fact, most of the time, the results are correct. The bad results are due to the limitations of the algorithm itself.

kuaashish commented 2 months ago

Hi @Javacr,

Thank you for bringing this issue to our attention. We are aware of the problem, as noted in this GitHub issue. For now, we are marking it as a bug and assigning it to our team for further investigation so that it will be on our radar However, we currently do not have a timeline for a fix and cannot provide any updates at this time.

Javacr commented 1 month ago

Hi @Javacr,

Thank you for bringing this issue to our attention. We are aware of the problem, as noted in this GitHub issue. For now, we are marking it as a bug and assigning it to our team for further investigation so that it will be on our radar However, we currently do not have a timeline for a fix and cannot provide any updates at this time.

Thank you, look forward to your future resolution of this bug.