Hi, apologies if this has been asked before or is documented.
Is there a rule of thumb for how small an image can be and still be tagged well by e.g. the general model? I know this is highly dependent on the image content, but if you have a recommendation for the common case let me know. In my case I'm dealing with photographs taken by users - no speciality subjects.
Similarly, is there a recommendation for minimum quality of lossy compression (e.g. JPEG)? Basically the same question along a separate access.
Naturally, I want to minimize the size of the images I send to Clarifai and also maximize the speed with which I generate them (hence lower JPEG quality when possible).