search

Arthur151 / ROMP

Monocular, One-stage, Regression of Multiple 3D People and their 3D positions & trajectories in camera & global coordinates. ROMP[ICCV21], BEV[CVPR22], TRACE[CVPR2023]
https://www.yusun.work/
Apache License 2.0
1.36k stars 231 forks source link

simple_romp #415

Open lllllialois opened 1 year ago

lllllialois commented 1 year ago

What's the difference between romp and simple_romp? My understanding is: romp is a project with more complete functions, including data preprocessing, training, testing and other functions. simple_romp takes the inference function out of the package and focuses more on an inference function. And I have some questions。

  1. It looks as if simple_ro is packaged as a software package. If I just wanted to use romp for reasoning, I could use simple_romp (Do I need to configure the environment that the full romp depends on? Or can you use it as a package?)

  2. If I want to retrain my own data set and use weights to reason, should I retrain with ROMP and then store the retrained weights as ROMP.pkl? Using simple_romp to infer?

  3. When I run simple_romp, there are operations to visualize the mesh, but if I want to store the mesh as obj, there don't seem to be any parameters to choose from. What should I do if I want to store the mesh? Is that right to modify https://github.com/Arthur151/ROMP/blob/f5b87be2769a70d5c2aba11f734fb48fc5bc9a23/simple_romp/romp/main.py? But simple_romp seems to be a package. Do I need to go to the site-packages to modify them?

    
# show the results during processing image / video, add:
--show

items you want to visualized, including mesh,pj2d,j3d,mesh_bird_view,mesh_side_view,center_conf,rotate_mesh. Please add if you want to see more:

--show_items=mesh,mesh_bird_view



4. If I later want to make some functional changes to romp for my tasks, it seems that the corresponding simple_romp will have to be changed, too?
Arthur151 commented 1 year ago

Thanks for your interests in simple-romp.

1.If you only need the inference part, then proper installaton of simple-romp is enough. You don't have to set the full environment for training.

2.Yes, you can use the trained model for simple-romp for inference. Just convert it and properly set the path of used checkpoint.

3.You can get all outputs from https://github.com/Arthur151/ROMP/blob/f5b87be2769a70d5c2aba11f734fb48fc5bc9a23/simple_romp/romp/main.py#L175 including the vertices and faces of mesh. If you modify the code, you just need to do this:

pip uninstall simple-romp
cd /path/to/modified/simple-romp
python setup.py install

Then you will use the modified code for inference during calling romp.

4.Yes.

lllllialois commented 1 year ago

thank u for ur reply！