Windsrain / Selective-Stereo

[CVPR 2024 Highlight] Selective-Stereo: Adaptive Frequency Information Selection for Stereo Matching
MIT License
115 stars 12 forks source link

关于论文中Middlebury训练的问题 #15

Closed iaphantom closed 4 months ago

iaphantom commented 4 months ago

请问下训练Middlebury数据集的时候,除了MiddEval3明确了分辨率,比如2005和2006,使用的full还是half分辨率的数据集?

Windsrain commented 4 months ago

请看stereo_datasets.py的fetch_dataloader部分,构建数据集时明确了分辨率是多少,默认是F。

iaphantom commented 4 months ago

非常感谢