Fix config activation passing to SPPBottleneck

[adamp87]

Bugfix: the SPPBottleneck in all three tinynas implementation is missing the passing of the activation function defined in the config file. This PR fixes it. See export of training logs at bottom.

This PR adds ReLU6 to ops.py as selectable activation function. Furthermore, it fixes the visualization for demo.py, as OpenCV fails with readonly numpy arrays. See, opencv/opencv#24522. This fixes #138 and obsoletes #160

Training log, branch main

        (5): SuperResStem(
          (act): ReLU6(inplace=True)
          (block_list): ModuleList(
            (0): MobileV3Block(
              (conv): Sequential(
                (0): Conv2d(160, 1760, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (1): SyncBatchNorm(1760, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (2): ReLU6(inplace=True)
                (3): Conv2d(1760, 1760, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=1760, bias=False)
                (4): SyncBatchNorm(1760, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (5): Identity()
                (6): ReLU6(inplace=True)
                (7): Conv2d(1760, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (8): SyncBatchNorm(320, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
              )
            )
            (1): SPPBottleneck(
              (conv1): ConvBNAct(
                (conv): Conv2d(320, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (bn): SyncBatchNorm(160, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (act): SiLU(inplace=True) # BUG: activation not passed
              )
              (m): ModuleList(
                (0): MaxPool2d(kernel_size=5, stride=1, padding=2, dilation=1, ceil_mode=False)
                (1): MaxPool2d(kernel_size=9, stride=1, padding=4, dilation=1, ceil_mode=False)
                (2): MaxPool2d(kernel_size=13, stride=1, padding=6, dilation=1, ceil_mode=False)
              )
              (conv2): ConvBNAct(
                (conv): Conv2d(640, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (bn): SyncBatchNorm(320, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (act): SiLU(inplace=True) # BUG: activation not passed
              )
            )
            (2): MobileV3Block(
              (conv): Sequential(
                (0): Conv2d(320, 1760, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (1): SyncBatchNorm(1760, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (2): ReLU6(inplace=True)
                (3): Conv2d(1760, 1760, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1760, bias=False)
                (4): SyncBatchNorm(1760, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (5): Identity()
                (6): ReLU6(inplace=True)
                (7): Conv2d(1760, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (8): SyncBatchNorm(320, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
              )
            )
          )
        )

Training log, this PR

        (5): SuperResStem(
          (act): ReLU6(inplace=True)
          (block_list): ModuleList(
            (0): MobileV3Block(
              (conv): Sequential(
                (0): Conv2d(160, 1760, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (1): SyncBatchNorm(1760, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (2): ReLU6(inplace=True)
                (3): Conv2d(1760, 1760, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=1760, bias=False)
                (4): SyncBatchNorm(1760, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (5): Identity()
                (6): ReLU6(inplace=True)
                (7): Conv2d(1760, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (8): SyncBatchNorm(320, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
              )
            )
            (1): SPPBottleneck(
              (conv1): ConvBNAct(
                (conv): Conv2d(320, 160, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (bn): SyncBatchNorm(160, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (act): ReLU6(inplace=True) # fixed
              )
              (m): ModuleList(
                (0): MaxPool2d(kernel_size=5, stride=1, padding=2, dilation=1, ceil_mode=False)
                (1): MaxPool2d(kernel_size=9, stride=1, padding=4, dilation=1, ceil_mode=False)
                (2): MaxPool2d(kernel_size=13, stride=1, padding=6, dilation=1, ceil_mode=False)
              )
              (conv2): ConvBNAct(
                (conv): Conv2d(640, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (bn): SyncBatchNorm(320, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (act): ReLU6(inplace=True) # fixed
              )
            )
            (2): MobileV3Block(
              (conv): Sequential(
                (0): Conv2d(320, 1760, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (1): SyncBatchNorm(1760, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (2): ReLU6(inplace=True)
                (3): Conv2d(1760, 1760, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1760, bias=False)
                (4): SyncBatchNorm(1760, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
                (5): Identity()
                (6): ReLU6(inplace=True)
                (7): Conv2d(1760, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)
                (8): SyncBatchNorm(320, eps=0.001, momentum=0.03, affine=True, track_running_stats=True)
              )
            )
          )
        )