-
Notifications
You must be signed in to change notification settings - Fork 39
Commit
This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.
Support different models per camera (#148)
- Loading branch information
Showing
33 changed files
with
861 additions
and
122 deletions.
There are no files selected for viewing
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,127 @@ | ||
| from pydantic import Field, validator, root_validator | ||
| from typing import List, Optional | ||
|
|
||
| from .base import SnakeModel | ||
|
|
||
|
|
||
| MODELS_DEVICES = { | ||
| "Jetson": ["ssd_mobilenet_v2_coco", "ssd_mobilenet_v2_pedestrian_softbio", "openpifpaf_tensorrt"], | ||
| "EdgeTPU": ["mobilenet_ssd_v2", "pedestrian_ssd_mobilenet_v2", "pedestrian_ssdlite_mobilenet_v2", "posenet"], | ||
| "Dummy": ["openvino", "openpifpaf_tensorrt", "mobilenet_ssd_v2", "openpifpaf", "yolov3", "ssd_mobilenet_v2_coco", | ||
| "ssd_mobilenet_v2_pedestrian_softbio", "posenet", "pedestrian_ssd_mobilenet_v2", | ||
| "pedestrian_ssdlite_mobilenet_v2"], # All available models. | ||
| "x86": ["mobilenet_ssd_v2", "openvino", "openpifpaf", "openpifpaf_tensorrt", "yolov3"], | ||
| "x86-gpu": ["mobilenet_ssd_v2", "openvino", "openpifpaf", "openpifpaf_tensorrt", "yolov3"] | ||
| } | ||
|
|
||
|
|
||
| class MLModelDTO(SnakeModel): | ||
| device: str = Field(example="Jetson") | ||
| name: str = Field(example="openvino") | ||
| image_size: str = Field(example="300,300,3") | ||
| model_path: str = Field("") | ||
| class_id: int = Field(1) | ||
| min_score: float = Field(0.25, example=0.30) | ||
| tensorrt_precision: Optional[int] = Field(None, example=32) | ||
|
|
||
| @validator("device") | ||
| def validate_device(cls, device): | ||
| if device not in ["x86", "x86-gpu", "Jetson", "EdgeTPU", "Dummy"]: | ||
| raise ValueError('Not valid Device. Try one of the following: "x86", "x86-gpu", "Jetson", "EdgeTPU",' | ||
| '"Dummy".') | ||
| return device | ||
|
|
||
| @validator("tensorrt_precision") | ||
| def validate_tensorrt_precision(cls, tensorrt_precision): | ||
| if tensorrt_precision is not None and tensorrt_precision not in [16, 32]: | ||
| raise ValueError('Not valid tensorrt_precision. Accepted values: 16, 32') | ||
|
|
||
| return tensorrt_precision | ||
|
|
||
| @validator("image_size") | ||
| def validate_image_size(cls, image_size): | ||
| integers = image_size.split(",") | ||
|
|
||
| if len(integers) != 3: | ||
| raise ValueError('ImageSize must be a string with 3 numbers separated with commas. Ex: "30,30,40".') | ||
|
|
||
| try: | ||
| [int(x) for x in integers] | ||
| except ValueError: | ||
| raise ValueError('ImageSize must be a string with 3 numbers separated with commas. Ex: "30,30,40".') | ||
|
|
||
| return image_size | ||
|
|
||
| # Root validators are called after each field validators success. | ||
|
|
||
| @root_validator(skip_on_failure=True) | ||
| def check_models_and_device(cls, values): | ||
| if values.get("device") == "Jetson": | ||
| if values.get("name") not in MODELS_DEVICES["Jetson"]: | ||
| raise ValueError(f'The device {values.get("device")} only supports the following models:' | ||
| f' {MODELS_DEVICES["Jetson"]}. ') | ||
|
|
||
| elif values.get("device") == "EdgeTPU": | ||
| if values.get("name") not in MODELS_DEVICES["EdgeTPU"]: | ||
| raise ValueError(f'The device {values.get("device")} only supports the following models:' | ||
| f' {MODELS_DEVICES["EdgeTPU"]}. ') | ||
|
|
||
| elif values.get("device") == "Dummy": | ||
| # No restrictions on this model. | ||
| # All available models. | ||
| if values.get("name") not in MODELS_DEVICES["Dummy"]: | ||
| raise ValueError(f'The device {values.get("device")} only supports the following models:' | ||
| f' {MODELS_DEVICES["Dummy"]}. ') | ||
|
|
||
| elif values.get("device") == "x86": | ||
| if values.get("name") not in MODELS_DEVICES["x86"]: | ||
| raise ValueError(f'The device {values.get("device")} only supports the following models:' | ||
| f' {MODELS_DEVICES["x86"]}. ') | ||
|
|
||
| elif values.get("device") == "x86-gpu": | ||
| if values.get("name") not in MODELS_DEVICES["x86-gpu"]: | ||
| raise ValueError(f'The device {values.get("device")} only supports the following models:' | ||
| f' {MODELS_DEVICES["x86-gpu"]}. ') | ||
|
|
||
| return values | ||
|
|
||
| @root_validator(skip_on_failure=True) | ||
| def check_variables_for_models(cls, values): | ||
|
|
||
| if values.get("name") in ["ssd_mobilenet_v2_coco", "ssd_mobilenet_v2_pedestrian_softbio", "mobilenet_ssd_v2", | ||
| "pedestrian_ssd_mobilenet_v2", "pedestrian_ssdlite_mobilenet_v2", "mobilenet_ssd_v2", | ||
| "openvino"]: | ||
| pass | ||
|
|
||
| elif values.get("name") == "openpifpaf_tensorrt": | ||
| if values.get("tensorrt_precision") is None: | ||
| raise ValueError('The model "openpifpaf_tensorrt" requires the parameter "tensorrt_precision", and' | ||
| 'said parameter can be either 16 or 32.') | ||
|
|
||
| integers = values.get("image_size").split(",") | ||
| if int(integers[0]) % 16 != 1 or int(integers[0]) % 16 != 1: | ||
| raise ValueError('First two values of ImageSize must be multiples of 16 plus 1 for openpifpaf_tensorrt.' | ||
| ' Ex: "641,369,3".') | ||
|
|
||
| elif values.get("name") == "posenet": | ||
| integers = values.get("image_size").split(",") | ||
|
|
||
| if integers != ["1281", "721", "3"] and integers != ["641", "481", "3"] and integers != ["481", "353", "3"]: | ||
| raise ValueError('ImageSize must be either one of the following options: "1281,721,3", "641,481,3" or' | ||
| ' "481,353,3".') | ||
|
|
||
| elif values.get("name") == "openpifpaf": | ||
| if values.get("image_size").split(",") != ["1281", "721", "3"]: | ||
| raise ValueError('ImageSize must be 1281,721,3 for the model "openpifpaf".') | ||
|
|
||
| elif values.get("name") == "yolov3": | ||
| integers = values.get("image_size").split(",") | ||
|
|
||
| reminder_one = int(integers[0]) % 32 | ||
| reminder_two = int(integers[1]) % 32 | ||
|
|
||
| if reminder_one or reminder_two != 0: | ||
| raise ValueError('For yolov3 model the ImageSize MUST be w = h = 32x e.g: x= 13=> ImageSize = 416,' | ||
| '416,3. ') | ||
|
|
||
| return values |
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
| Original file line number | Diff line number | Diff line change |
|---|---|---|
| @@ -0,0 +1,70 @@ | ||
| import json | ||
| import os | ||
| from pathlib import Path | ||
|
|
||
|
|
||
| from fastapi import APIRouter, status, Body | ||
| from starlette.exceptions import HTTPException | ||
| from typing import Optional | ||
|
|
||
| from api.models.ml_model import MLModelDTO | ||
| from api.routers.cameras import validate_camera_existence | ||
| from api.settings import Settings | ||
| from api.utils import restart_processor, handle_response | ||
| from libs.utils.config import get_source_config_directory | ||
|
|
||
|
|
||
| ml_model_router = APIRouter() | ||
| settings = Settings() | ||
|
|
||
|
|
||
| def pascal_case_to_snake_case(parameters): | ||
| result = {} | ||
|
|
||
| for key, value in parameters: | ||
| result[''.join(word.title() for word in key.split('_'))] = value | ||
|
|
||
| if "ClassId" in result.keys(): | ||
| result["ClassID"] = result["ClassId"] | ||
| del result["ClassId"] | ||
|
|
||
| return result | ||
|
|
||
|
|
||
| @ml_model_router.post("/{camera_id}") | ||
| async def modify_ml_model(camera_id: str, model_parameters: MLModelDTO, reboot_processor: Optional[bool] = True): | ||
| validate_camera_existence(camera_id) | ||
|
|
||
| parameters = pascal_case_to_snake_case(model_parameters) | ||
|
|
||
| base_path = os.path.join(get_source_config_directory(settings.config), camera_id) | ||
| models_directory_path = os.path.join(base_path, "ml_models") | ||
| json_file_path = os.path.join(models_directory_path, f"model_{parameters['Device']}.json") | ||
|
|
||
| model_name = parameters["Name"] | ||
| del parameters["Name"] | ||
| del parameters["Device"] | ||
|
|
||
| # Create or modify .json file | ||
| json_content = { | ||
| "model_name": model_name, | ||
| "variables": parameters | ||
| } | ||
|
|
||
| if os.path.exists(json_file_path): | ||
| with open(json_file_path, 'w') as outfile: | ||
| json.dump(json_content, outfile) | ||
| else: | ||
| # Hypothesis: source config directory (base_path) should always exists. | ||
| if not os.path.exists(models_directory_path): | ||
| Path(models_directory_path).mkdir(parents=True, exist_ok=True) | ||
|
|
||
| with open(json_file_path, 'x+') as outfile: | ||
| json.dump(json_content, outfile) | ||
|
|
||
| # Reboot processor if set | ||
| success = True | ||
| if reboot_processor: | ||
| success = restart_processor() | ||
|
|
||
| return handle_response(json_content, success, decamelize=False) |
Oops, something went wrong.