search

open-mmlab / mmsegmentation

OpenMMLab Semantic Segmentation Toolbox and Benchmark.
https://mmsegmentation.readthedocs.io/en/main/
Apache License 2.0
8.29k stars 2.62k forks source link

what does "img_norm_cfg" mean? #793

Closed ghost closed 3 years ago

ghost commented 3 years ago

Hi, thank you for your code, I found that the following code appears in different datasets, what does it mean? img_norm_cfg = dict(mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)

If I use a custom dataset, should I modify it? Actually, in addition to RGB channels, my dataset also has infrared band and DEM, it means one image has 5 channels, What should I pay attention to when customizing the dataset?

MengzhangLI commented 3 years ago

Hi, @wujf98:

The img_nor_cfg is image normalization configuration which is used for image normalization before being inputed into network. It is related with our used pretrained models (most of them are based on ImageNet) and pretrained models are very important for model training.

Thus, if you also want to use pretrained model, please keep the img_norm_cfg identical.

Right now, MMSegmentation just supports RGB input, we would consider adding more features for other modalities or input images.

Best,

ghost commented 3 years ago

Thanks a lot, @MengzhangLI