Object Classification, Detection and Tracking with Rigorous Maths Behind
We are mastering the usage of Dlib and Caffe Frameworks with C++ and C++ only.
We are new to deep learning but we are serious about it. We have deep expertise in typical machine learning algorithms, adaptive filtering, step size and leak adaptations in LMS and backpropagation neural networks. We are at state of the art level in terms of tracking skill set as well. Therefore for us it is not difficult to catch up with the top-notch developments in the field. Just like any one else skilled in the art we see the future of computer vision inside deep learning.
Our Three Musketeers in Deep Learning
Deep learning frameworks are just tools. As long as one does not learn rigorous mathematics and essential algorithmic principles behind, they are alone useless.
One has to be proficient in statistical signal processing and adaptive filtering in order to do something useful with deep learning. Otherwise you'll be nothing but another tool of the eco-system.
cuDNN & Caffe
Caffe is the mostly used CNN (convolutional neural network) framework. cuDNN is the Nvidia library which makes life easier in deep learning.
Needless to tell GPU and CUDA are musts in deep learning.
Dlib is not used as frequently as Caffe or TensorFlow by the community but it does the job well. It is 100% C++, it does not get in your way and hence we love it; it is simply our most favorite.
Bored of drinking Caffe all day then try Dlib. It might a bit shake you off and rotate ypur world but worth it.
Supports cuDNN as well, of course.