tfliteをFlaskで動かしてクライアントからのリクエストで起動させる

# work on Flask server
# server will be activated only when client request is occured
#
#
from flask import Flask, jsonify
import cv2
import numpy as np
import tflite_runtime.interpreter as tflite
from picamera2 import Picamera2
from waitress import serve

app = Flask(__name__)

# モデル・ラベル初期化
interpreter = tflite.Interpreter(model_path="efficientdet_lite0.tflite")
interpreter.allocate_tensors()
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
with open("coco_labels.txt", "r") as f:
    labels = [line.strip() for line in f.readlines()]

# カメラ初期化
picam2 = Picamera2()
picam2.preview_configuration.main.size = (640, 480)
picam2.preview_configuration.main.format = "RGB888"
picam2.preview_configuration.align()
picam2.configure("preview")
picam2.start()

def preprocess_image(image):
    resized = cv2.resize(image, (320, 320))
    resized = resized[:, :, [2, 1, 0]]  # BGR→RGB
    return np.expand_dims(resized, axis=0).astype(np.uint8)

def postprocess_results(boxes, scores, classes, count, image_shape, labels):
    detections = []
    for i in range(count):
        if scores[i] > 0.4:
            ymin, xmin, ymax, xmax = boxes[i]
            left, right, top, bottom = (
                int(xmin * image_shape[1]),
                int(xmax * image_shape[1]),
                int(ymin * image_shape[0]),
                int(ymax * image_shape[0])
            )
            detections.append({
                'box': [left, top, right, bottom],
                'class_id': int(classes[i]),
                'score': float(scores[i]),
                'label': labels[int(classes[i])] if int(classes[i]) < len(labels) else f"id:{int(classes[i])}"
            })
    return detections

def detect_once():
    frame = picam2.capture_array()
    input_data = preprocess_image(frame)
    interpreter.set_tensor(input_details[0]['index'], input_data)
    interpreter.invoke()
    boxes = interpreter.get_tensor(output_details[0]['index'])[0]
    classes = interpreter.get_tensor(output_details[1]['index'])[0]
    scores = interpreter.get_tensor(output_details[2]['index'])[0]
    count = int(interpreter.get_tensor(output_details[3]['index'])[0])
    detections = postprocess_results(boxes, scores, classes, count, frame.shape, labels)
    return detections

@app.route('/detect', methods=['GET'])
def detect_route():
    detections = detect_once()
    return jsonify(detections)

if __name__ == '__main__':
	# when you limit access only from the local machine, use a loopback address instead of 0.0.0.0
    app.run(host='0.0.0.0', port=5000)

$ python tflite_flask.py
INFO: Created TensorFlow Lite XNNPACK delegate for CPU.
[0:43:36.818382413] [87808]  INFO Camera camera_manager.cpp:327 libcamera v0.4.0+53-29156679
[0:43:36.825774869] [87830]  INFO RPI pisp.cpp:720 libpisp version v1.1.0 e7974a156008 27-01-2025 (21:50:51)
[0:43:36.914007444] [87830]  INFO RPI pisp.cpp:1179 Registered camera /base/axi/pcie@1000120000/rp1/i2c@88000/imx708@1a to CFE device /dev/media0 and ISP device /dev/media1 using PiSP variant BCM2712_C0
[0:43:36.916831533] [87808]  INFO Camera camera.cpp:1202 configuring streams: (0) 640x480-RGB888 (1) 1536x864-BGGR_PISP_COMP1
[0:43:36.916945996] [87830]  INFO RPI pisp.cpp:1484 Sensor: /base/axi/pcie@1000120000/rp1/i2c@88000/imx708@1a - Selected sensor format: 1536x864-SBGGR10_1X10 - Selected CFE format: 1536x864-PC1B
 * Serving Flask app 'tflite_flask'
 * Debug mode: off
WARNING: This is a development server. Do not use it in a production deployment. Use a production WSGI server instead.
 * Running on all addresses (0.0.0.0)
 * Running on http://127.0.0.1:5000
 * Running on http://192.168.1.19:5000
Press CTRL+C to quit
192.168.1.18 - - [26/Apr/2025 19:16:35] "GET /detect HTTP/1.1" 200 -

import requests

if __name__ == '__main__':
    result = requests.get('http://rasp5.local:5000/detect').json()
    print(result)

<人がカメラから見える位置>
% python tfreq.py
[{'box': [3, -1, 636, 481], 'class_id': 0, 'label': 'person', 'score': 0.9375}]


<体をカメラからずらすと>
% python tfreq.py
[{'box': [143, 23, 607, 472], 'class_id': 81, 'label': 'id:81', 'score': 0.4765625}]