Learning to Estimate Indoor Lighting from 3D Objects

Project page

Explanation video

Demonstration

---

Dependencies:

• pytorch (https://pytorch.org/)
• Install the following lib/toolbox (also python setup.py install):
  • sudo apt-get install openexr libopenexr-dev
  • skylibs (https://github.com/weberhen/skylibs.git)
  • pytorch_toolbox (https://github.com/weberhen/pytorch_toolbox)

Installation

python setup.py install

Downloading the models/dataset

• Link to the pre-trained illumination predictor models (183MB).

• Link to the LDR datasets to train the illumination predictor. (3.5GB)

• place the models at /learning_indoor_lighting/IlluminationPredictor/models. Ex: models/bun_zipper_glossy/model_best.pth.tar

• place the datasets at /learning_indoor_lighting/Datasets. Ex: Datasets/indoor_hdr/train and Datasets/objects_ldr/bun_zipper_glossy/train

Visualizing training/testing

Activate visdom in another terminal:

python -m visdom.server

Then go to the terminal: localhost:8097

Testing the autoencoder

cd learning_indoor_lighting/AutoEncoder && python test.py

Testing the illumination predictor

cd learning_indoor_lighting/IlluminationPredictor && python test.py

Citation

@inproceedings{weber_3dv_18,
  author    = {Henrique Weber and
               Donald Pr{\'{e}}vost and
               Jean{-}Fran{\c{c}}ois Lalonde},
  title     = {Learning to Estimate Indoor Lighting from 3D Objects},
  booktitle = {International Conference on 3D Vision, Verona, Italy},
  pages     = {199--207},
  year      = {2018},
}