Continuous integration and tensorflow_cc

[PeterBowman]

PR https://github.com/roboticslab-uc3m/vision/pull/93 lacked CI support on Travis due to build timeouts (https://github.com/roboticslab-uc3m/vision/pull/93#issuecomment-465135839), it takes way too long to compile this beast from sources.

Perhaps this is the perfect scenario for docker (https://github.com/roboticslab-uc3m/vision/pull/93#issuecomment-485287192 and next comments). If we could prepare a docker image in which every dependency, including tensorflow_cc, is installed and then pulled by Travis, build times should be considerably shorter. Use https://hub.docker.com/u/roboticslabuc3m to host said images (or rather docker run files, I'm still a docker rookie).

[davidvelascogarcia]

Maybe using tensorflow_cc docker image. I´ve been doing some tests in tensorflowDetection2D repo using dockerimage. The configuration that i used was .travis.yml. Using that configuration the script build and install yarp from source and install openCV. Required time take around 15-20 minutes by commit. Attached travis CI example using docker.

[jgvictores]

Dropping as wontfix in favor of https://github.com/roboticslab-uc3m/vision/issues/103, right @PeterBowman ?

[PeterBowman]

I think we should keep docker on our radar because of GPU/DNN-ready OpenCV 4.x, but yes, this issue should be dropped as we aim to supersede tensorflowDetection2D. Closing!