Transform normalization
I have a question regarding the normalization done in the transformation. Namely, if I look at the config file it mentions "mean": [0.485, 0.456, 0.406 ], "std": [0.229, 0.224, 0.225 ].
However, when I look at the transform from
transforms = create_transform(**resolve_data_config(model.pretrained_cfg, model=model)),
all means and stds are 0.5. So, which ones to use?
Also, what type of stain normalization is recommended? Macenko?
"mean": [0.485, 0.456, 0.406 ], "std": [0.229, 0.224, 0.225 ]
These are the correct values to use.
I'm not able to reproduce what you are encountering with 0.5 mean/std values. Could you share what version of timm you are using? And also could you run print(model.pretrained_cfg) and share the output?
Thanks!
I now also see these values when I print(model.pretrained_cfg).