Reference to #1255

Fix https://github.com/matterport/Mask_RCNN/issues/870

Fix visualize activations (Squashed 4 commits by @keineahnung2345)

# Get activations of a few sample layers
activations = model.run_graph([image], [
    ("input_image",        model.keras_model.get_layer("input_image").output)
])
leads to the error:
InvalidArgumentError: input_image:0 is both fed and fetched.

Revise the code according to https://stackoverflow.com/questions/39307108/placeholder-20-is-both-fed-and-fetched

Fix typo "ceate" to "create"

In compute_overlaps_masks(), the code checking the shape of masks1 and masks2 is not correct.
1. It should check masks' last dimension instead of first dimension to determine whether it's empty or not
2. When masks1 or masks2 is empty, it should return a zero array of shape (masks1.shape[-1], masks2.shape[-1])

Valid index 0 is not regarded, leading to possibly more prediction_matches than grountruh_matches and recall values higher than 1.

In current version of log(), when you pass a zero size array, it will throw an error:
`ValueError: Unknown format code 'f' for object of type 'str'`.
This PR fixes it.

pass name=None to tf.boolean_mask will lead to ValueError: At least one of name (None) and default_name (None) must be provided.

Fix the input shape of anchors in ProposalLayer and use "num_anchors" to replace "anchors" to avoid confusing

* fix comment in DetectionTargetLayer
* fix comment of target_deltas' shape
* make parenthesis compatible

skimage generates different warnings depending on the version. This wrapper function calls skimage.tranform.resize() with the right parameter for each version.

1. fix description on window
2. fix output shape of detection layer

As discussed in: https://github.com/matterport/Mask_RCNN/issues/919

1. fix the comment on output shape in fpn_classifier_graph
2. unify NUM_CLASSES and num_classes to NUM_CLASSES
3. unify boxes, num_boxes, num_rois, roi_count to num_rois
4. use more specific POOL_SIZE and MASK_ POOL_SIZE to replace pool_height and pool_width

For PyramidROIAlign's output shape, use pool_height and pool_width instead of height and width to avoid confusion with those of feature_maps.

add IMAGE_CHANNEL_COUNT class variable to config to make it easier to use Mask_RCNN for non 3-channel images

Fixed comment on GPU_COUNT

Fix bugs of not showing boxes in 1.b RPN Predictions.
TF 1.9 introduces "ROI/rpn_non_max_suppression/NonMaxSuppressionV3:0", so the original code can't work.

Unified regex expression for both, Linux and Windows.

- Current implementation creates self.log_dir in set_log_dir() function,
  which creates an empty log directory if none exists. This causes
  find_last() to fail after creating a model because it finds this new
  empty directory instead of the previous training directory.
- New implementation moves log_dir creation to the train() function to
  ensure it is only created when it will be used.

VIA has changed JSON formatting in later versions. Now instead of a dictionary, "regions" has a list, see the issue #928