Flask服务推理图片返回结果

import traceback
import base64
import tensorflow as tf
import cv2
import numpy as np
from flask import Flask, request, jsonify
from tensorflow.python.saved_model import tag_constants
import collections

# initiate
app = Flask(__name__)
iou_thresh = 0.5
score_thresh = 0.3

#load_model
model = tf.saved_model.load("./futi/yolov4-416", tags=[tag_constants.SERVING]) #用yolov4的模型，将模型转换成pb格式
infer = model.signatures['serving_default']


def bbox2points(bbox):
    ymin, xmin, ymax, xmax = bbox
    
    return ymin,xmin, ymax, xmax

def detector(image):
    boxes_filtered = []
    scores_filtered = []
    box_ids_filtered = []
    image = cv2.resize(image, (608, 608))
    image = image/255.
    image_data = []
    image_data.append(image)
    image_data = tf.constant(np.asarray(image_data).astype(np.float32))
    pred_bbox = infer(image_data)
    for key, value in pred_bbox.items():
        boxes = value[:, :, 0:4]
        pred_conf = value[:, :, 4:]
    boxes, scores, classes, valid_detections = tf.image.combined_non_max_suppression(
        boxes=tf.reshape(boxes, (tf.shape(boxes)[0], -1, 1, 4)),
        scores=tf.reshape(
            pred_conf, (tf.shape(pred_conf)[0], -1, tf.shape(pred_conf)[-1])),
        max_output_size_per_class=100,
        max_total_size=200,
        iou_threshold=iou_thresh,
        score_threshold=score_thresh
    )
    # print(scores,boxes,NUM_CLASS)
    pred_bbox = [boxes.numpy(), scores.numpy(), classes.numpy(), valid_detections.numpy()]
    boxes, scores, classes, count = pred_bbox
    classes = classes[0].tolist()
    scores = scores[0].tolist()
    boxes = boxes[0].tolist()
    count = count[0]
    for i in range(count):
        boxes_filtered.append(bbox2points(boxes[i]))
        scores_filtered.append(scores[i])
        box_ids_filtered.append(int(classes[i])+1)

    res = {"outputs":{"detection_boxes": [boxes_filtered], 
    "detection_score:": [scores_filtered], "detection_classes": [box_ids_filtered]}}
    return res

#predict & post
@app.route('/predict',methods = ['POST'])
def post_data_ladder():
    # get data
    try:
        print("[INFO]: Enacting api..")
        base64img = None
#get_json(force=True)设置force=true解决 Nonetype can not suscripble的问题
        data = request.get_json(force=True) if request.method == 'POST' else request.arg.get('name')  
        base64img = data["inputs"][0]["b64"]
        print('[INFO]: Image receive successfully')
        if base64img is None:
            print("[ERROR]: Image cant be None")
            return jsonify(None)
        # Inference
        img = base64.b64decode(base64img)
        np_array = np.fromstring(img, np.uint8)
        img_np = cv2.imdecode(np_array, cv2.IMREAD_COLOR)
        detect_img = cv2.cvtColor(img_np, cv2.COLOR_BGR2RGB)
        result = detector(detect_img)
    except Exception as e:
        traceback.print_exc()
        print(repr(e))
        print('Analysis.Detect fail')
        return jsonify(None)
    return jsonify(result)

if __name__ == '__main__':
    app.run(host='0.0.0.0', port=8501)   #设置host127.0.0.1或者0.0.0.0

参考资料：
https://pytorch.apachecn.org/docs/1.4/28.html
https://blog.csdn.net/weixin_42902669/article/details/84590282
https://blog.csdn.net/newbeixue/article/details/103482399