Hi @michaeltrs, first thanks for the interesting paper and the well documented codebase!
I want to see if I can reproduce your experiments on object classification on PASTIS-24. In the paper and supplementary material it is mentioned that:
To make the PASTIS classification dataset we took advantage of the object instance ids provided to extract 24 × 24 pixel regions whose center pixel falls inside each object and use the class of this object as the sample class.
Although I do see the experiment of classification_train_transf.py and there is a TSViT_cls.yaml configuration, it points to PASTIS24_cls dataset, which does not exist in datasets.yaml. So far I understood that get_pastis_dataloader is used for classification as well.
Would you have the exact code available that you used to generate PASTIS24-cls? I assume it is a variant of data2windows.py.
Thanks!
Hi @michaeltrs, first thanks for the interesting paper and the well documented codebase!
I want to see if I can reproduce your experiments on object classification on PASTIS-24. In the paper and supplementary material it is mentioned that:
To make the PASTIS classification dataset we took advantage of the object instance ids provided to extract 24 × 24 pixel regions whose center pixel falls inside each object and use the class of this object as the sample class.Although I do see the experiment of
classification_train_transf.pyand there is aTSViT_cls.yamlconfiguration, it points toPASTIS24_clsdataset, which does not exist indatasets.yaml. So far I understood thatget_pastis_dataloaderis used for classification as well.Would you have the exact code available that you used to generate PASTIS24-cls? I assume it is a variant of
data2windows.py.Thanks!