(from https://www.cityscapes-dataset.com/)
Ailia input shape: (512, 1024, 3) RGB order Range: [0, 255] 8-bit unsigned integer
Automatically downloads the tflite file on the first run. It is necessary to be connected to the Internet while downloading.
For the sample image,
$ python3 hrnet_segmentation.py
If you want to specify the input image, put the image path after the --input option. You can use --savepath option to change the name of the output file to save.
$ python3 hrnet_segmentation.py --input IMAGE_PATH --savepath SAVE_IMAGE_PATH
By adding the --video option, you can input the video. If you pass 0 as an argument to VIDEO_PATH, you can use the webcam input instead of the video file.
$ python3 hrnet_segmentation.py --video VIDEO_PATH --savepath SAVE_VIDEO_PATH
We have three pretrained-model:
- HRNetV2-W48
- HRNetV2-W18-Small-v1
- HRNetV2-W18-Small-v2 (default)
You can specify the architecture you want following --arch / -a option.
$ python3 hrnet_segmentation.py --arch HRNetV2-W48
Two versions of the model are provided: full integer quantization (8-bit) and full precision floating point (32-bit). By default, the full integer quantization is used but the user can select the other version by passing the --float flag.
$ python3 hrnet_segmentation.py --float
https://github.com/HRNet/HRNet-Semantic-Segmentation
Tensorflow 2.11.0