Which dataset has been used to pre-train the vision models?


in order to evaluate the performance of different models (not only pre-trained models), I want to use datasets from ImageNet.
The problem is, that I want to go sure, that the images I present, are ‘new’ to the model in order to control the bias but it could be very likely, that popular datasets were used to train the models.
The papers related to the models only mention the datasets used to benchmark the models.
Is there a way to get informations about how these networks have been trained?

Thank you very much in advance.

The classification models were pretrained on ImageNet, while the segmentation and detection models use e.g. COCO. The pretrained argument gives you information about the used dataset.

1 Like