workflows / app.py
SkalskiP's picture
debug
0547b86
raw history blame
No virus
7.01 kB
import json
from typing import List
import cv2
import os
import numpy as np
import gradio as gr
import supervision as sv
from inference_sdk import (
InferenceHTTPClient,
InferenceConfiguration,
VisualisationResponseFormat
)
def read_json_file(file_path: str) -> dict:
with open(file_path, 'r') as file:
return json.load(file)
def split_and_strip(text: str) -> List[str]:
return [part.strip() for part in text.split(',')]
MARKDOWN = """
# WORKFLOWS 🛠
Define complex ML pipelines in JSON and execute it, running multiple models, fusing
outputs seamlessly.
Use self-hosted Inference HTTP [container](https://inference.roboflow.com/inference_helpers/inference_cli/#inference-server-start)
or run against Roboflow [API](https://detect.roboflow.com/docs)
to get results without single line of code written.
"""
# LICENSE PLATES WORKFLOW
LICENSE_PLATES_MARKDOWN = """
![license-plates-detection-workflow](
https://media.roboflow.com/inference/license-plates-detection-workflow.png)
"""
LICENSE_PLATES_EXAMPLES = [
"https://media.roboflow.com/inference/license_plate_1.jpg",
"https://media.roboflow.com/inference/license_plate_2.jpg",
]
LICENSE_PLATES_SPECIFICATION_PATH = 'configs/license_plates.json'
LICENSE_PLATES_SPECIFICATION = read_json_file(LICENSE_PLATES_SPECIFICATION_PATH)
LICENSE_PLATES_SPECIFICATION_STRING = f"""
```json
{json.dumps(LICENSE_PLATES_SPECIFICATION, indent=4)}
```
"""
# CAR BRAND WORKFLOW
CAR_BRANDS_MARKDOWN = """
![car-brand-workflow](
https://media.roboflow.com/inference/car-brand-workflow.png.png)
"""
CAR_BRANDS_EXAMPLES = [
["Lexus, Honda, Seat", "https://media.roboflow.com/inference/multiple_cars_1.jpg"],
["Volkswagen, Renault, Mercedes", "https://media.roboflow.com/inference/multiple_cars_2.jpg"],
]
CAR_BRANDS_SPECIFICATION_PATH = 'configs/car_brands.json'
CAR_BRANDS_SPECIFICATION = read_json_file(CAR_BRANDS_SPECIFICATION_PATH)
CAR_BRANDS_SPECIFICATION_STRING = f"""
```json
{json.dumps(CAR_BRANDS_SPECIFICATION, indent=4)}
```
"""
API_URL = os.getenv('API_URL', None)
API_KEY = os.getenv('API_KEY', None)
print("API_URL", API_URL)
if API_KEY is None or API_URL is None:
raise ValueError("API_URL and API_KEY environment variables are required")
CLIENT = InferenceHTTPClient(api_url=API_URL, api_key=API_KEY)
CLIENT.configure(InferenceConfiguration(
output_visualisation_format=VisualisationResponseFormat.NUMPY))
def annotate_image(image: np.ndarray, detections: sv.Detections) -> np.ndarray:
h, w, _ = image.shape
annotated_image = cv2.cvtColor(image, cv2.COLOR_RGB2BGR)
line_thickness = sv.calculate_dynamic_line_thickness(resolution_wh=(w, h))
text_scale = sv.calculate_dynamic_text_scale(resolution_wh=(w, h))
bounding_box_annotator = sv.BoundingBoxAnnotator(thickness=line_thickness)
label_annotator = sv.LabelAnnotator(
text_scale=text_scale,
text_thickness=line_thickness
)
annotated_image = bounding_box_annotator.annotate(
annotated_image, detections)
annotated_image = label_annotator.annotate(
annotated_image, detections)
return cv2.cvtColor(annotated_image, cv2.COLOR_BGR2RGB)
def inference_license_plates(input_image: np.ndarray) -> np.ndarray:
result = CLIENT.infer_from_workflow(
specification=LICENSE_PLATES_SPECIFICATION["specification"],
images={"image": input_image},
)
detections = sv.Detections.from_inference(result)
if len(detections) == 0:
return input_image
detections['class_name'] = (
result["recognised_plates"]
if isinstance(result["recognised_plates"], list)
else [result["recognised_plates"]]
)
return annotate_image(input_image, detections)
def inference_car_brands(input_text: str, input_image: np.ndarray) -> np.ndarray:
classes = split_and_strip(input_text)
result = CLIENT.infer_from_workflow(
specification=CAR_BRANDS_SPECIFICATION["specification"],
images={"image": input_image},
parameters={"car_types": classes}
)
detections = sv.Detections.from_inference(result)
if len(detections) == 0:
return input_image
if len(detections) > 1:
class_ids = np.argmax(result["clip"], axis=1)
else:
class_ids = np.array([np.argmax(result["clip"], axis=0)])
detections.class_ids = class_ids
detections['class_name'] = [classes[class_id] for class_id in class_ids]
return annotate_image(input_image, detections)
with gr.Blocks() as demo:
gr.Markdown(MARKDOWN)
with gr.Tab(label="License Plates"):
gr.Markdown(LICENSE_PLATES_MARKDOWN)
with gr.Accordion("Configuration JSON", open=False):
gr.Markdown(LICENSE_PLATES_SPECIFICATION_STRING)
with gr.Row():
license_plates_input_image_component = gr.Image(
type='numpy',
label='Input Image'
)
license_plates_output_image_component = gr.Image(
type='numpy',
label='Output Image'
)
with gr.Row():
license_plates_submit_button_component = gr.Button('Submit')
gr.Examples(
fn=inference_license_plates,
examples=LICENSE_PLATES_EXAMPLES,
inputs=license_plates_input_image_component,
outputs=license_plates_output_image_component,
cache_examples=True
)
with gr.Tab(label="Car Brands"):
gr.Markdown(CAR_BRANDS_MARKDOWN)
with gr.Accordion("Configuration JSON", open=False):
gr.Markdown(CAR_BRANDS_SPECIFICATION_STRING)
with gr.Row():
with gr.Column():
car_brands_input_image_component = gr.Image(
type='numpy',
label='Input Image'
)
car_brands_input_text = gr.Textbox(
label='Car Brands',
placeholder='Enter car brands separated by comma'
)
car_brands_output_image_component = gr.Image(
type='numpy',
label='Output Image'
)
with gr.Row():
car_brands_submit_button_component = gr.Button('Submit')
gr.Examples(
fn=inference_car_brands,
examples=CAR_BRANDS_EXAMPLES,
inputs=[car_brands_input_text, car_brands_input_image_component],
outputs=car_brands_output_image_component,
cache_examples=True
)
license_plates_submit_button_component.click(
fn=inference_license_plates,
inputs=license_plates_input_image_component,
outputs=license_plates_output_image_component
)
car_brands_submit_button_component.click(
fn=inference_car_brands,
inputs=[car_brands_input_text, car_brands_input_image_component],
outputs=car_brands_output_image_component
)
demo.launch(debug=False, show_error=True)