DanNet achieved the first deep convolutional neural network wins across four computer vision contests from May 2011 to September 2012, researcher Lucas Beyer notes, before ImageNet became the dominant benchmark

Strategically bad choice of competitions.

They were all minor competitions only a very small subset of the community cared (or even heard) about.

On the other hand, ImageNet was considered a bit of a grand challenge and very hard but general benchmark.

Another one they could have done is Pascal VOC, then the community would have cared too, but it was too small to make any DL work.