Coarse-to-Fine Vision-Language Pre-training with Fusion in the Backbone - View it on GitHub