For image data, Caffe (http://caffe.berkeleyvision.org/ http://caffe.berkeleyvision.org/) is dominant (most deep learning computer vision research I see is done

For image data, Caffe (http://caffe.berkeleyvision.org/ http://caffe.berkeleyvision.org/) is dominant (most deep learning computer vision research I see is done with Caffe). However, that's the chief problem with Caffe - it's very difficult to extend because it revolves so much around "just specify attributes" and train.