Flow-to-depth (FDNet) video-depth-estimation

This is the implementation of paper

Video Depth Estimation by Fusing Flow-to-Depth Proposals

Jiaxin Xie, Chenyang Lei, Zhuwen Li, Li Erran Li, Qifeng Chen

In IROS 2020.

See our paper (https://arxiv.org/pdf/1912.12874.pdf) for more details. Please contact Jiaxin Xie (jxieax@ust.cse.hk) if you have any questions.

Prerequisites

This codebase was developed and tested with Tensorflow 1.4.0 and Numpy 1.16.2

Evaluation on KITTI Eigen Split

IF you want to generate GroundTruth Depth from KITTI RAW data, download KITTI dataset using this script provided on the official website.

Meanwhile, we also provided GroundTruth Depth save in npy file, download it from here

Our final results on KITTI Eigen is availible on here

Then run

python kitti_eval/eval_depth_general.py --kitti_dir=/path/to/raw/kitti/dataset/ or /path/to/downloaded/GoundTruth/npy/file/ --pred_file=/path/to/our/final/results/