scott-mao/DIFFNet

Self-Supervised Monocular Depth Estimation with Internal Feature Fusion, BMVC2021

DIFFNet

This repo is for Self-Supervised Monocular Depth Estimation with Internal Feature Fusion(arXiv), BMVC2021

A new backbone for self-supervised depth estimation.

• Unlike other lead competitors, only test-time refinement (TTR) is applied on 1024x320 monocular model, no supervisory signals are involved with this rank.

If you think it is not a bad work, please consider citing it.

@inproceedings{diffnet_bmvc,
    title = {Self-Supervised Monocular Depth Estimation with Internal Feature Fusion},
    author  = {Hang Zhou, David Greenwood and Sarah Taylor},
    booktitle = {British Machine Vision Conference (BMVC)},
    month = {November},
    year = {2021}}

Comparing with others

Evaluation on selected hard cases:

Trained weights

• diffnet_640x192
• diffnet_640x192_ms
• diffnet_1024x320
• diffnet_1024x320_ttr

Setting up before training and testing

• Data preparation: please refer to monodepth2

Training:

sh start2train.sh

• Note:

Testing:

sh disp_evaluation.sh

Infer a single depth map from a RGB:

sh test_sample.sh

Acknowledgement

Thanks the authors for their works:

• monodepth2
• HRNet