How does Bag of words perform under scale and rotation changes? Are scale and rotation the only metrics or are there any other metrics checked for evaluation of general computer vision method?
In computer vision, the bag-of-words model (BoW model) can be applied to image classification, by treating image features as words.
In other words, you just detect all features you can find in the image and treat them as unordered set.
Whether these features are invariant to scaling, translation, rotation hence completely depends only on the features you use.