Mask-ToF

This is the code for the CVPR 2021 work: Mask-ToF: Learning Microlens Masks for Flying Pixel Correction in Time-of-Flight Imaging

If you use bits and bobs of this code, or find inspiration from it, consider citing the paper:

@article{chugunov2021masktof,
title={Mask-ToF: Learning Microlens Masks for Flying Pixel Correction in Time-of-Flight Imaging},
author={Chugunov, Ilya and Baek, Seung-Hwan and Fu, Qiang and Heidrich, Wolfgang and Heide, Felix},
journal={The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
year={2021}
}

Requirements:

• Developed using PyTorch 1.7.0 on Ubuntu x64 machine
• Condensed requirements in /requirements.txt
• Full frozen environment can be found in /conda_env.txt, but many of these libraries are not necessary to run this code

Data:

• Download test and additional from https://lightfield-analysis.uni-konstanz.de/
• Place samples directly in /data
• Confirm that the .txts in /filenames match the data structure (update these if you add/reorganize data)
• Optional: Download pretrained network checkpoints from this drive link
• Extract pretrained.zip into \nets\pretrained
• Optional: Download 'barcode' masks from this drive link
• Extract barcode_masks.zip into \utils\barcode_masks

Project Structure:

MaskToF
  ├── checkpoints  
  │   └── // folder for network checkpoints
  ├── data  
  │   └── // folder for training/test data
  ├── dataloader  
  │   ├── dataloader.py  // pytorch dataloader for lightfields + depth
  │   └── transforms.py  // data augmentations and code to generate image/depth patches
  ├── filenames   
  │   └── // .txt files pointing to data locations
  ├── model.py  // wrapper class for training the network:
  │             // -> load data, calculate loss, print to tensorboard, save network state
  ├── nets  
  │   ├── MaskToFNet.py  // the meat of MaskToF, class for data simulation and learning a mask:
  │   │                  // -> simulate measurements, pass to refinement network, return depth
  │   ├── pretrained
  │   │   └── // folder for pretrained networks
  │   └── refinement  
  │       └── // refinement network architectures
  ├── train.py  // wrapper class for arg parsing and setting up training loop
  └── utils  
      ├── chamfer_distance 
      │   └── // pytorch implementation of chamfer distance metric
      ├── file_io.py  // utils for loading light field data
      ├── tof.py  // utils for simulating time-of-flight measurements
      └── utils.py  // miscellaneous helper functions (e.g. saving network state)

Training:

• Should be as simple as running the debug example with bash debug.sh, if all the prerequisite libraries play nice
• Outputs will be saved to checkpoint_dir=checkpoints/debug/
• --init mask_pattern_name sets the initial mask iterate, in this case gaussian_circles1.5,0.75, which is the Gaussian Circle pattern with mean 1.5 and standard distribution 0.75
• --use_net is a flag to jointly train a refinement/reconstruction network, as outlined in the paper
• Additional arguments and descriptions can be found at the top of /train.py

Reconstruction:

The notebook reconstruction.ipynb includes an interactive demo for loading a network from a checkpoint folder, visualizing mask structure, simulating amplitude measurements, and performing depth reconstruction.

Experimental Setup:

If you're building your own experimental prototype, you can reach out to me at chugunov[at]princeton[dot]edu for information and advice.

Best of luck,
Ilya