search

zxz267 / HaMuCo

[ICCV 2023] HaMuCo: Hand Pose Estimation via Multiview Collaborative Self-Supervised Learning
https://zxz267.github.io/HaMuCo/
MIT License
44 stars 4 forks source link

Thank you for your reply and hard work! #4

Closed qwer904 closed 7 months ago

qwer904 commented 8 months ago

Thank you for your reply and hard work! I have two more questions to ask you. 1.How can I input a picture and then display the hand skeleton? like in the paper. 2.Can the input be changed from a picture to real-time gesture recognition using a depth camera?

zxz267 commented 8 months ago

Thank you for your reply and hard work! I have two more questions to ask you. 1.How can I input a picture and then display the hand skeleton? like in the paper. 2.Can the input be changed from a picture to real-time gesture recognition using a depth camera?

  1. You can structure the input image to match the format expected by the "process_single_view_input" function. Subsequently, proceed to pass it through the model to obtain the output, which should then undergo processing via the "process_output" function. Finally, utilize this function to display the 2D skeleton, utilizing the first two dimensions of the "output["coord_uvd"]" data.
  2. I believe the input of HaMuCo could be altered to accommodate depth images by converting them into 3 channels. However, HaMuCo does not fully account for the nature of depth images, rendering it suboptimal for this input format and potentially necessitating further adjustments.