grapeaaa / sanet

Improving Monocular Depth Estimation by Leveraging Structural Awareness and Complementary Datasets

This repository contains code to compute depth from a single image. It accompanies our paper:

Improving Monocular Depth Estimation by Leveraging Structural Awareness and Complementary Datasets
Tian Chen⋆, Shijie An⋆, Yuan Zhang, Chongyang Ma, Huayan Wang, Xiaoyan Guo, and Wen Zheng

Setup

1. Download the model weights (full model training on NYUv2 dataset) SANet-NYUv2.pth(password:x55o) and place the file in the pretrained folder.

2. Set up dependencies:

   conda install pytorch torchvision opencv

   The code was tested with Python 3.6, PyTorch 1.1.0, and OpenCV 4.1.2.

Usage

1. Place the pretrained model in the folder pretrained.

2. Run the model:

   python test.py --cuda

Results

(1) Our network architecture

(2) Our result compare on NYUv2 dataset

Citation

Please cite our paper if you use this code or any of the models:

@article{Tian2020,
	author    = {Tian Chen⋆, Shijie An⋆, Yuan Zhang, Chongyang Ma, Huayan Wang, Xiaoyan Guo, and Wen Zheng},
	title     = {Improving Monocular Depth Estimation by Leveraging Structural Awareness and Complementary Datasets},
	journal   = {},
	year      = {2020},
}

License

MIT License