search

VCIP-RGBD / DFormer

[ICLR 2024] DFormer: Rethinking RGBD Representation Learning for Semantic Segmentation
https://yinbow.github.io/Projects/DFormer/index.html
MIT License
142 stars 24 forks source link

When using raw depth maps，is this done by copying the single-channel depth image input 3 times？ #12

Closed kikyou2018 closed 6 months ago

kikyou2018 commented 6 months ago

在RGBXDataset.py文件找到对应的处理
https://github.com/VCIP-RGBD/DFormer/blob/main/utils/dataloader/RGBXDataset.py#L52 这段代码是继承自cmx项目，关于dformer的处理 https://github.com/VCIP-RGBD/DFormer/blob/main/models/encoders/DFormer.py#L282 可以看出，简单复制三次后，又仅取一个通道。