mirror of
https://github.com/blakeblackshear/frigate.git
synced 2026-02-14 15:15:22 +03:00
d360c4f808
This change migrates the frigate container build to use Debian trixie as the base image. This permits us to use newer upstream packages (and, for example, stop needing to use a custom ffmpeg build). The main hitch was the need for Python 3.9 for Pycoral from the Google apt repository, for Coral Edge TPU support. Fortunately, the open source community has stepped up, and there are now TFLite and Pycoral wheels available for Python 3.10-3.12 as well.
20 lines
818 B
Python
20 lines
818 B
Python
import openvino as ov
|
|
from openvino.tools import mo
|
|
from pathlib import Path
|
|
from site import getsitepackages
|
|
|
|
def transformations_config_location(path: Path):
|
|
return path / 'openvino' / 'tools' / 'mo' / 'front' / 'tf' / 'ssd_v2_support.json'
|
|
|
|
configs = [transformations_config_location(Path(path)) for path in getsitepackages()]
|
|
assert len(configs) > 0, 'Expected at least one transformations config to exist but none existed.'
|
|
|
|
ov_model = mo.convert_model(
|
|
"/models/ssdlite_mobilenet_v2_coco_2018_05_09/frozen_inference_graph.pb",
|
|
compress_to_fp16=True,
|
|
transformations_config=configs[0],
|
|
tensorflow_object_detection_api_pipeline_config="/models/ssdlite_mobilenet_v2_coco_2018_05_09/pipeline.config",
|
|
reverse_input_channels=True,
|
|
)
|
|
ov.save_model(ov_model, "/models/ssdlite_mobilenet_v2.xml")
|