From b4ba44f570b177446cbfa674c5f05fbdf59f11e9 Mon Sep 17 00:00:00 2001 From: Timur Bolotyuh Date: Thu, 3 Aug 2023 13:31:54 +0200 Subject: [PATCH] Update README-classification-tutorial.md When you want fine-tuning model on a custom dataset with another N of classes you have to pass argument `--model.resume-exclude-scopes classifier` otherwise, we will get an error about the mismatch classifier layer. --- .../en/models/classification/README-classification-tutorial.md | 1 + 1 file changed, 1 insertion(+) diff --git a/docs/source/en/models/classification/README-classification-tutorial.md b/docs/source/en/models/classification/README-classification-tutorial.md index 6ef4c08..de223c2 100644 --- a/docs/source/en/models/classification/README-classification-tutorial.md +++ b/docs/source/en/models/classification/README-classification-tutorial.md @@ -51,6 +51,7 @@ If we want to finetune the ImageNet model with `1000` classes on another classif * Pass this argument `--model.classification.finetune-pretrained-model` to enable finetuning * Specify number of classes in pre-trained model using `--model.classification.n-pretrained-classes` argument * Specify the location of pre-trained weights using `--model.classification.pretrained` argument + * Pass this argument `--model.resume-exclude-scopes classifier` For a concrete example, see training recipe of [MobileViTv2](README-mobilevit-v2.md)