search

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
798 stars 90 forks source link

Configuration files #8

Closed neon5d closed 4 years ago

neon5d commented 4 years ago

Hello, Thank you for sharing your great work. I wonder whether you can share configuration files of ShapeNet and NMR for single_view_reconstruction/single(multi)_view_supervision and multi_view_reconstruction.

There are some configurations but I am not sure it can be applied for other datasets and supervision. Thank you,

m-niemeyer commented 4 years ago

Hi @neon5d , thanks for your interest! Can you please specify which config you think is missing? Because as you mentioned, there are already respective config files in the folders, e.g. single_view_reconstruction/multi_view_supervision/ours_depth.yaml can be used to train a single-view reconstruction model with depth supervision. Thanks!

neon5d commented 4 years ago

It seems "single_view_reconstruction/multi_view_supervision/ours_rgb.yaml" is configured for NMR_Dataset Can I use the same configuration for ShapeNet after changing "data->path" to derive papers' results?

For "single_view_reconstruction/multi_view_supervision/ours_rgb.yaml" and "single_view_reconstruction/multi_view_supervision/combined_rgb.yaml" , Are these config files can be used for ShapeNet and NMR to derive your paper's experiment?

If I need some changes other than "data->path", can you share your configuration files of ShapeNet and NMR for the paper's experiment? Thank you

m-niemeyer commented 4 years ago

Hi @neon5d , OK, I see what you mean - thanks for the clarification!

First, "ShapeNet" and "NMR" are both the ShapeNet dataset! Only the renderings of the objects are different. We did this because most 3D supervised methods use the Choy et. al. renderings, but most 2D supervised methods use the renderings from Kato et. al. Please see the "Dataset" paragraph under section 4.1 on page 6 of the paper for more details.

Hence, for all experiments from the paper, there is a .yaml config file and a _pretrained.yaml config file. Hence, you do not need to change the path! I hope this helps, otherwise please let me know!

neon5d commented 4 years ago

Thank you for clarification.