Hi
I use clara_xray_classification_chest_amp as a pre-trained model.
I finetune the model using train_finetune.sh in the project with a new dataset (different class numbers).
After I called the finetune script, I got exceptions as attached screenshots.
I’m not sure this issue related to Clara SDK version or not because the model was trained in Clara SDK 2.0 but I finetune it using Clara SDK 3.0.
Hi
In order to help you better. Could you share the top part of the data list.json specially the "label_format": section.
Also are you able to run train.sh ? just wondering if the error is only when running the finetune.sh
Thanks
Hi
the top part of the datalist.json
{“label_format”: [1],
“training”: [
{“image”: “14.png”, “label”: [0]},
{“image”: “15.png”, “label”: [0]},
and yes, I can run train.sh.
Thanks
Hi
It looks like you are doing a binary classification. In order to use the pre-trained model you would still need to keep the same format with 15 labels. unfortunatly we are unable to break the head and train a new head. So you can simply ignore 14 out of 15 labels so it would be
{“label_format”: [1,1,1,1,1,1,1,1,1,1,1,1,1,1,1],
“training”: [
{“image”: “14.png”, “label”: [0,0,0,0,0,0,0,0,0,0,0,0,0,0,0]}, --> neg class
{“image”: “15.png”, “label”: [1,0,0,0,0,0,0,0,0,0,0,0,0,0,0]}, --> pos class
Also depending on what is the disease you are targeting and its location picking different indexes may give you different results.
FYI you should modify the class index in the metric section of the config.json
Hope that helps
ok, i see.
Thanks for your help.