vcl3d/deeppanoramalighting

Deep Lighting Environment Map Estimation from Spherical Panoramas (CVPRW20)

Code accompanying the paper "Deep Lighting Environment Map Estimation from Spherical Panoramas", CVPRW 2020

TODO

• Pre-trained model.
• Inference code.

Code and Trained Models

This repository contains inference code and models for the paper Deep Lighting Environment Map Estimation from Spherical Panoramas (link).

Requirements

The code is based on PyTorch and has been tested with Python 3.7 and CUDA 10.0. We recommend setting up a virtual environment (follow the virtualenv documentation) for installing PyTorch and the other necessary Python packages. Once your environment is set up and activated, install the necessary packages:

pip install torch===1.2.0 torchvision===0.4.0 -f https://download.pytorch.org/whl/torch_stable.html

Inference

You can download pre-trained models from here, which includes pre-trained LDR-to-HDR autoencoder and Lighting Encoder. Please put the extracted files under models and run:

python inference.py

The following flags specify the required parameters.

• --input_path: Specifies the path of the input image.
• --out_path: Specifies the file of the output path.
• --deringing: Enable/disable low pass deringing filter for the predicted SH coefficients.

deeppanoramalighting's Issues

About sh distribution prior

The paper mentioned that thanks to the use of distributed priors, the performance of the network has been greatly improved.
But the code does not have an implementation of Equation.10. The last layer of the illumination model is a fully connected layer, and the Equation.1 is not used.
Can you introduce the specific usage of Equation.10? Which part of the code does it work? Thank you!

the training code

hi , woluld you release the training code?

thanks

Pre-trained model link error

The link address for downloading the pre-trained model is incorrect.
Please check the link.

How can I get an HDRI image output?

If I run the code, I get this

I need a HDRI image like this:

Any help will be much appreciated...

about the resolution of image

When I change the size of image，I will get the error? Does the size of the input image have to be resized to 512 × 256 ？

How to get spherical harmonics Coeeficients by sh_functions.py

Hi !

I am trying to find codes to help me compute spherical harmonics Coeeficients from hdr panorama images(such as panoramas from LavalInDoor datasets). I have read your paper and code, and i think the function "getCoeeficientsMatrix(...)" in sh_functions.py can help me, but i haven't found how to set the image as a input in this function.
Can you tell me how to use sh_functions.py to compute spherical harmonics Coeeficients? Or if sh_functions.py couldn't do this, can anyone tell my which code can do this?

Thanks!