Training for a single class , single image reconstruction

autonomousvision / differentiable_volumetric_rendering

This repository contains the code for the CVPR 2020 paper "Differentiable Volumetric Rendering: Learning Implicit 3D Representations without 3D Supervision"

http://www.cvlibs.net/publications/Niemeyer2020CVPR.pdf

MIT License

794 stars 90 forks source link

Training for a single class , single image reconstruction #47

Closed trilokpadhi closed 3 years ago

trilokpadhi commented 3 years ago

Hi Thanks for the nice repo !!

I have prepared a dataset for a single class , also written a script to render its views in blender from its corresponding 3d object so I have image-object pair (approx 900).

Now next can you please tell how to prepare the data to load to the model and train on custom data ?

Thanks

m-niemeyer commented 3 years ago

Please have a look at the readme on how to train a new model.

Our data structure has the format:

[ds_folder] [class_0] [class_1] ... [class_n] [obj_0] [0bj_1] ... [obj_m]

Each "obj_m" folder contains "camera.npz" and the subfolders "image", "mask", "visual_hull_depth" and/or "depth". You can see how the data is loaded and processed in the main image field.

Good luck with your research!