
TODO:
    TrainingModes:
        [x] categorical
            see demos on:
                [x] MNIST
                [x] Cifar100
                [ ] ImageNet
                [ ] ...
        [ ] segmentation
            [ ] semantic
                [X] CamVid
                [ ] CityScapes
                [ ] Diva
                [ ] UrbanMapper3D
                [ ] ...
            [ ] instance
                [ ] UrbanMapper3D
        [ ] tracking
            [ ] ...
        [ ] detection
            [ ] ...
            [X] VOC2007
        [ ] identification
            [ ] 1-vs-all
            [ ] N-vs-all
            [X] (1-vs-1) pairwise
            [ ] (N-vs-N)
            [ ] ...
