TensorFlow or Caffe: Which is better for running Deep learning convolutional neural networks
TensorFlow and Caffe cannot be compared on the basis of performance as both of these do not show a significant difference. Although when it comes to the language you are using, you can define which of these work better.
Also, it is more important to focus on the hyper-parameters, than the tool you use to work with.
@AishwaryaSingh has a valid point!
I’d like to give a historical perspective of the two libraries in terms of the “vision and mission” behind building them.
Caffe was built specifically to deploy Deep Learning models to production. Right now its successor Caffe2 is the most dependable library in the industry for end-to-end deployment. On the other hand, TensorFlow was built for RnD of deep learning in terms of how they can be used to enhance the performance of full fledged products. Although I would say that a lot of efforts are being put to make TensorFlow a library for everything related to building deep learning products; whether it be creating new models from scratch to deploying models in production.
So to summarize, I still would prefer Caffe2 for production as of now - and wait for TensorFlow to become the best library for everything.