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

How to test this on new image? #19

Closed nitish11 closed 4 years ago

nitish11 commented 4 years ago

How to generate 3D model from a single 2D image?

The config file mentioned below does't contain parameter test_input image path. https://github.com/autonomousvision/differentiable_volumetric_rendering/blob/master/configs/single_view_reconstruction/single_view_supervision/ours_combined_pretrained.yaml

Which part of the code generate 3D model, how to change the output format of 3D model?

m-niemeyer commented 4 years ago

Hi @nitish11 ,

if you only want to run our pretrained models on RGB images, please have a look at our demo. You can further adjust the path to the image in the respective demo config file. Finally, in the default config you can find the config entry

generation:
  mesh_extension: ply

which you can use to change the output format

nitish11 commented 4 years ago

Thanks @m-niemeyer for the hint. I was able to generate good quality of 3D models from random test images just by changing config file, configs/demo/demo_combined.yaml. 

inherit_from: configs/single_view_reconstruction/multi_view_supervision/ours_combined_pretrained.yaml
data:
  dataset_name: images
  path: media/test_images
training:
  out_dir:  media/out_3D_models
generation:
  generation_dir: generation

media/test_images : contains all the images media/out_3D_models : contains all 3D models.