search

google-ai-edge / mediapipe

Cross-platform, customizable ML solutions for live and streaming media.
https://ai.google.dev/edge/mediapipe
Apache License 2.0
27.83k stars 5.18k forks source link

Question about using pose_landmark_full_body.tflite #1914

Closed WayneLin015 closed 3 years ago

WayneLin015 commented 3 years ago

Different to the BlazePose paper, it outputs (ld_3d) with the shape of [1,195], in my opinion and previous issue, it should be flatten refer to x, y, z, visibility, presence, so the landmark result will be 39 outputs though its different to paper said that the model output 33 landmark

Another question is, I need to apply any other post or pre processing? Now the output seem there's some shifting The result of 39 points tt (2) With python API result

maokk commented 3 years ago

I have the same problem on pose_landmark_upper_body.tflite file but works ok on python API, and I find the tflite outputs value in ld_3d are not normalized to [0~1]

Python API annotated_image tflite file output my_testimage

maokk commented 3 years ago

anything update? thanks

chaowentao commented 3 years ago

I have the same problem on pose_landmark_upper_body.tflite file but works ok on python API, and I find the tflite outputs value in ld_3d are not normalized to [0~1]

Python API annotated_image tflite file output my_testimage

@maokk @WayneLin015 How do you test the tflite model ? Can you give me the run.py? Now I am solving this problem. Thank you ~

chaowentao commented 3 years ago

@WayneLin015 The results contain 39 points, 1-33 is the pose landmark, 34-39 is auxiliary points for tracking. My tflite result is bad. So can you share your inference code? Thank you~

sgowroji commented 3 years ago

Hi @WayneLin015, Could you please provide the steps to reproduce the above observation mentioned in the first comment. Thanks!

google-ml-butler[bot] commented 3 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you.

google-ml-butler[bot] commented 3 years ago

Closing as stale. Please reopen if you'd like to work on this further.