/GW-Depth

Primary LanguagePython

Monocular Depth Estimation for Glass Walls with Context: A New Dataset and Method

example
We captured a RGB-D dataset for real world glass walls. A glass structure context and reflection context combined method is proposed to facilitate the depth estimation for glass walls. Our method demonstrates effective for both glass walls and indoor scenes in NYU Depth V2.

Training

bash script/train/a0_train_stage1_res50_glassrgbd.sh ckpt_0

ckpt_0 denotes the directory to save the trained checkpoints.

Evaluation

bash script/evaluation/eval_stage1_line-glassrgbd.sh ckpt_0