SweepNet

This repository contains Matlab codes for paper, "SweepNet:Wide-baseline Omnidirectional Depth Estimation" (ICRA 2019). [arxiv]

Contact: Changhee Won ([email protected])

Prerequisites

List of code/library dependencies

• MATLAB R2017b, 2018a, 2018b, 2019a
• CUDA 9.0, 10.0
• MatConvNet 1.0-beta25

How to run

Compile mex files (c++, cuda)

>> compile
(output:
  ./mex/build/mexConcurrentSGM.mex*
  ./mex/build/mexGPUSGM.mex*
  ./mex/build/mexInterp2D.mex*
  ./mex/build/mexCUDAInterp2D.mex*)

Test (run_test_sweepnet.m)

• Pretrained weights: [sweepnet_sunny_14.mat]

• Set arguments in the script

  • "matconvnet_path": path to matconvnet
  • "pretrained_model": path to pretrained_weights
  • "data_opts.db_path": path to dataset

• Run

>> run_test_sweepnet

Example result

Dataset

You can download the synthetic datasets in the project page.

The directory structure should be like this:

sunny/
     /cam1/
     /cam2/
     /cam3/
     /cam4/
     /depth_train_640/
     /intrinsic_extrinsic.mat

Citation

Will be updated soon