[WIP][bashscript/android] Support tflite segmentation model

[niklasjang]

It is a sub-issue of #62 Support tensorflow-lite basic model with bash_script .

As mentioned in style transfer reallocate issue, there is a realloc problem with style transfer model. So I am going to try to support segmentation example.

I am considering this segmentation model which is supported by tensorflow.

• Input: content image (257, 257, 3)
• Output: type: float32 | dimen: (257,257,21) // a probability each of 21 labels

//supported labels
LABEL_NAMES = np.asarray([
    'background', 'aeroplane', 'bicycle', 'bird', 'boat', 'bottle', 'bus',
    'car', 'cat', 'chair', 'cow', 'diningtable', 'dog', 'horse', 'motorbike',
    'person', 'pottedplant', 'sheep', 'sofa', 'train', 'tv'
])

Brief pipeline

gst-launch-1.0 -v \
      filesrc location=cat.jpg ! jpegdec ! videoconvert ! videoscale ! imagefreeze !\
      video/x-raw,format=RGB,width=257,height=257,framerate=10/1 ! tee name=t \
      t. tensor_converter !\
      tensor_transform mode=arithmetic option=typecast:float32,add:0.0,div:255.0 !\
      tensor_filter framework=tensorflow-lite model=deeplabv3_257_mv_gpu.tflite !\
      tensor_decoder mode=segmentation ! compositor name=mix sink_0::zorder=2 sink_1::zorder=1 !\
      videoconvert ! ximagesink
      t. ! queue leaky=2 ! mix.

At first, I will make decoder(output array -> image type) for this model and then construct bash script example. If it is done, I will expand it to android application example.

[taos-ci]

:octocat: cibot: Thank you for posting issue #69. The person in charge will reply soon.

[niklasjang]

This feature is merged at (https://github.com/nnsuite/nnstreamer/pull/1801).

I'm closing this.