The rules state that external data cannot be used. Does this extend to pre-trained models such as Inception, VGG or other image classification models which have information from external data implicitly embedded in their trained weights? Is their use allowed or forbidden?
In that same vein, can a pretrained model be used for feature engineering of the images? Specifically, can I use a object detection and localization model I’ve trained on another dataset to identify object categories and crop out rest of the image? I’m exploring treating this as both a detection and localization problem as well as a multi-class classification problem.