Open tony-hj opened 4 months ago
The readme mentions that this work can use a model-free setting and a small number of reference images to perform pose recognition. How can this be achieved?
you can refer to here for model-free setup. When you run on your own object, you just make your data the same format as the example.
this requires you to get poses for the ref images yourself. Another simpler way is to use bundlesdf. It will auto compute poses and mesh.
@wenbowen123 Thanks for you answer. I have make the dataset on my own object, I got rgb image , Depth and mask. depth_enhanced could get from depth and mask, But I have no idea how to get cam_in_ob? It seems like the last one i need to work.This is what i make. This your work need. Many place are different, Could you tell me how to make it like your work need.
I am now making a similar Linmod dataset format to this tutorial: https://gitee.com/Tong--Yuan/6-do-f_-linemod_datasets_making/raw/master/%E4%BD%8D%E5%A7%BF%E8%A7%A3%E7%AE%97%E6%95%B0%E6%8D%AE%E9%9B%86-Linemod%E6%95%B0%E6%8D%AE%E9%9B%86%E5%88%B6%E4%BD%9C%E6%96%B9%E6%B3%95(Linemod_preprocessed).pdf
@tony-hj Do you know how to get cam_in_ob? Using bundleSDF is slow and the model will be incomplete
@jingchenlu I'm sorry,I haven't fixed this yet
Thanks for your excellent work, I have no experience in posture detection. I want to apply this great method to my robot. How can I deploy it?run_demo.py can already work in my docker.Think you very much!!