Skip to content
This repository has been archived by the owner on Apr 8, 2024. It is now read-only.

Marker movement tracking #1

Closed
Rancho-zhao opened this issue Oct 28, 2022 · 3 comments
Closed

Marker movement tracking #1

Rancho-zhao opened this issue Oct 28, 2022 · 3 comments

Comments

@Rancho-zhao
Copy link

Hello! I have successfully used your codes to generate depth maps from tactile images. Next, I want to try marker movement tracking by myself, but I don't know how to build the dataset. If I use the supervised learning method, how can I label the marker movement in the tactile images of my dataset?

@vocdex
Copy link
Owner

vocdex commented Oct 28, 2022

Hello! Glad to hear it this repo was useful.
I just realized I haven't pushed the code for marker tracking. Until I do so, you can take a look at this repo from Gelsight:

  1. Optical flow method
    It works for DIGIT as well. You have to change the configs (trial&error) in **setting.py. My code is the same but for DIGIT.
    2.Neural Network method
    More recent, robust version of marker tracking. I had trouble with generating dataset for this model, as well.
    I'll let you know if I have updates with this.

@Rancho-zhao
Copy link
Author

Thank you for your reply! Your suggestions are very useful! Actually, I have used the neural network method in the repo you mentioned, but there are some problems I find. For example, the flow arrows are all in the upper left corner of the image when using the trained model to test. The reason I think may be the “generate_data” module . I'm trying to solve these problems.

@vocdex
Copy link
Owner

vocdex commented Oct 28, 2022

I guess you have to train a new model for DIGIT. Their provided generic model didn't work for our DIGIT, as well. And for optical flow method, it would run best if you cover the markers with complete black since it detects the markers based on HSV color histogram. Good luck!

Sign up for free to subscribe to this conversation on GitHub. Already have an account? Sign in.
Labels
None yet
Projects
None yet
Development

No branches or pull requests

2 participants