1) The biggest reason is label diversity. COCO is a great dataset, but at 90 labels you're going to be missing some potential. The problem with commercial machi

1) The biggest reason is label diversity. COCO is a great dataset, but at 90 labels you're going to be missing some potential. The problem with commercial machine learning development is access to data, and with Open Images and it's 545 classification labels we get that diversity. 2) Scaling is hard, Cross-language Interoperability is hard, And bringing your product to market is hard. Algorithmia solves all of those problems with grace and has a development experience that is second-to-none. Don't take my word for it, compare it against tensorflow-serving other competitor technologies. 3) I haven't yet but I could certainly see an improvement in performance, we did notice a dramatic difference between our SSD and faster-rcnn model as was expected. This was only an intro, but if you create a model that performs better than ours, we'd love to host it on Algorithmia.