Pytorch implementation for 'Interactive Reconstruction of Monte Carlo Image Sequences using a Recurrent Denoising Autoencoder' , https://research.nvidia.com/publication/interactive-reconstruction-monte-carlo-image-sequences-using-recurrent-denoising
Link to original paper (SIGGRAPH '17) : https://research.nvidia.com/publication/interactive-reconstruction-monte-carlo-image-sequences-using-recurrent-denoising
This is the unofficial PyTorch implementation of the above paper.
The input to this network is :
The output of this network is :
Construct the input as one image, as follows :
Column 1 | Column 2 |
---|---|
1 spp input | 250 spp output |
Albedo Image | Normal Map |
Depth Map | Roughness Map |
Note : The data directory must contain 'train' and 'test' directories, and these directories much contain directories where the sequence is stored.
To train the network, run the following command :
python train.py --data_dir [PATH_TO_DATA_DIRECTORY] --name [EXP_NAME] --save_dir [PATH_TO_SAVE_CHECKPOINTS] --epochs 500
To test the network, run the following command :
python test.py --data_dir [PATH_TO_DATA_DIRECTORY] --output_dir [PATH_TO_SAVE_RESULTS] --checkpoint [PATH_TO_CHECKPOINT].pt