deeplab_edgetpu2

DeepLab V3+ EdgeTPUV2 and AutoSeg EdgeTPU

Models

• EdgeTPU-optimized Vision Models - Semantic segmentation task
• autoseg-edgetpu
• deeplab-edgetpu

Environment

• Edge TPU Compiler version 16.0.384591198
• Google Coral EdgeTPU (run EdgeTPU Model)
• TensorFlow Lite (v2.5 or higher)
• OpenCV (3.x or 4.x)

Python Example.

Video capture

python3 deeplab_tflite_capture_opencv.py --help
usage: deeplab_tflite_capture_opencv.py [-h] --model MODEL [--width WIDTH] [--height HEIGHT] [--thread THREAD] [--videopath VIDEOPATH] [--output OUTPUT]

options:
  -h, --help            show this help message and exit
  --model MODEL         File path of Tflite model.
  --width WIDTH         Resolution width.
  --height HEIGHT       Resolution height.
  --thread THREAD       Num threads.
  --videopath VIDEOPATH
                        File path of Videofile.
  --output OUTPUT       File path of result.

Image

python3 deeplab_tflite_image_opencv.py --help
usage: deeplab_tflite_image_opencv.py [-h] --model MODEL [--input_shape INPUT_SHAPE] [--thread THREAD] [--input INPUT] [--output OUTPUT]

options:
  -h, --help            show this help message and exit
  --model MODEL         File path of Tflite model.
  --input_shape INPUT_SHAPE
                        Specify an input shape for inference.
  --thread THREAD       Num threads.
  --input INPUT         File path of image.
  --output OUTPUT       File path of result.

YouTube video Link

Coral Dev Board EdgeTPU AutosegEdgeTPU-XS with fusing argmax

Coral Dev Board EdgeTPU AutosegEdgeTPU-S with fusing argmax