I recommend you build using a pretrained Tensorflow model for your application. Training costs will be too high if you try to train your own models and there’s no point in reinventing the wheel (that’s the point of Crowdbotics!).
Looks like you’re making an image captioning application. I recommend you use one of these two open-source models:
The first one will work out of the box with Crowdbotics; the second one requires you to run some code first. Both of these will plug in nicely into our standard generated Django app. Over time we’ll add these as hot-pluggable modules like our others.