Upgrade to CUDA 11.2 and improve GPU support (#215)

* Updated to CUDA11.2

* Switch to devel image

* Update install.sh

* Update dependencies.ini

* Updated pytorch

* DCNv2 update

* Detectron update

* Update installation.md

* More strict stable-baselines3 ver

* Updated dep version

* Update CHANGELOG.md

* Small consistent fix

* Minor typo fix

* Updated CPU/GPU support for pip

* Updated numba version

* Removed OPENDR_DEVICE

* Added support for changing inference device during testing

* Style fixes

* Separate tests for face detection

* Updated testing pipeline for face detection 2d

* Update installation.md

* Fixed weight placement

* Fixed cuda placement check

* Fixed mxnet installation

* Updated base image

* Separate object detection 2d tests

* Upgraded to pytorch 1.9.0

* Changed default constant folding setting

* Updated EfficientPS

* Apply suggestions from code review

Co-authored-by: ad-daniel <44834743+ad-daniel@users.noreply.github.com>

* Bugfixes

* Increase tollerance in test_cox3d_learner

* Test dockerfile with correct branch

* Removed comment

* remove branch from docker files

Co-authored-by: ad-daniel <daniel.dias@epfl.ch>
Co-authored-by: ad-daniel <44834743+ad-daniel@users.noreply.github.com>
Co-authored-by: LukasHedegaard <lh@eng.au.dk>