classner
commented
3 years ago Hi @Aparajit-Garg!
The direct 3d prediction is the 'fast' version of the body fitting. This is described in the last part of the method section in the paper. It requires training decision forests to do this prediction. If you want to use it (and I'm not sure that that's the case) you first have to train the forests which generates the missing file, then you can use them.
If you 'just' want to fit the SMPL model to the points in the most accurate way, you can use the methods in the 3dfit folder.
Let me know how it goes!
Hi @classner, thank you for your quick response on my earlier query. I got stuck into another one, could you please help me out.
I am running direct3d/bodyfit.py after generating the pose from pose/pose.py from a .png image with height 513 and maintaining the aspect ratio. I'm passing a folder with a single image in it. The problem I'm facing is this:
No such file or directory: 'direct3d/../models/2dto3d/separate_regressors/forest_82_final.z'
and another thing is that 2 files are being generated namely, 1.) 33.png_body_directseparate.pkl_working 2.) work_lock
Upon opening these files the first one has the character 't' in it and the second one is empty.
Basically my end goal is to feed in an image and get a 3d model of it based on the 91 keypoints. I'm unable to figure out how to achieve this.
Can you please guide me with few details. I am running out of time here😅