When wiil you release the train code of seg?

mit-han-lab / efficientvit

EfficientViT is a new family of vision models for efficient high-resolution vision.

Apache License 2.0

1.59k stars 141 forks source link

Open Anivie opened 1 month ago

Anivie commented 1 month ago

Thanks for the excellent work! When wiil you release the train code of seg like sam?

fefespn commented 1 month ago

attached simple train script.

something to look at:

Change the preprocess and transformation if you wish for the dataset.
depending on your model of choose, change this line: model = create_seg_model("b2", "cityscapes", pretrained=False)
then in efficientvit/models/efficientvit/seg.py change the model you choose with the n_classes you have in your dataset.

(I am not from the repo contributors so it's not formal)