实例分割-Mask R-CNN 模型

本案例我们将进行实例分割模型Mask R-CNN的训练和测试的学习。在计算机视觉领域，实例分割(Instance Segmentation)是指从图像中识别物体的各个实例，并逐个将实例进行像素级标注的任务。实例分割技术在自动驾驶、医学影像、高精度GIS识别、3D建模辅助等领域有广泛的应用。本案例将对实例分割领域经典的Mask R-CNN模型进行简单介绍，并使用Matterport开源Mask R-CNN实现,展示如何在华为云ModelArts上训练Mask R-CNN模型。

点击跳转至Mask R-CNN模型详解

注意事项：

1. 本案例使用框架**:** TensorFlow-1.13.1

2. 本案例使用硬件规格**:** 8 vCPU + 64 GiB + 1 x Tesla V100-PCIE-32GB

3. 进入运行环境方法：点此链接进入AI Gallery，点击Run in ModelArts按钮进入ModelArts运行环境，如需使用GPU,您可以在ModelArts JupyterLab运行界面右边的工作区进行切换

4. 运行代码方法**:** 点击本页面顶部菜单栏的三角形运行按钮或按Ctrl+Enter键 运行每个方块中的代码

5. JupyterLab的详细用法**:** 请参考《ModelAtrs JupyterLab使用指导》

6. 碰到问题的解决办法**:** 请参考《ModelAtrs JupyterLab常见问题解决办法》

1.首先进行包的安装与引用

!pip install pycocotools==2.0.0

Collecting pycocotools==2.0.0
  Downloading http://repo.myhuaweicloud.com/repository/pypi/packages/96/84/9a07b1095fd8555ba3f3d519517c8743c2554a245f9476e5e39869f948d2/pycocotools-2.0.0.tar.gz (1.5MB)
    100% |████████████████████████████████| 1.5MB 52.3MB/s ta 0:00:01
[?25hBuilding wheels for collected packages: pycocotools
  Running setup.py bdist_wheel for pycocotools ... [?25ldone
[?25h  Stored in directory: /home/ma-user/.cache/pip/wheels/63/72/9e/bac3d3e23f6b04351d200fa892351da57f0e68c7aeec0b1b08
Successfully built pycocotools
Installing collected packages: pycocotools
Successfully installed pycocotools-2.0.0
You are using pip version 9.0.1, however version 21.0.1 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.

!pip install imgaug==0.2.9

Collecting imgaug==0.2.9
  Downloading http://repo.myhuaweicloud.com/repository/pypi/packages/17/a9/36de8c0e1ffb2d86f871cac60e5caa910cbbdb5f4741df5ef856c47f4445/imgaug-0.2.9-py2.py3-none-any.whl (753kB)
    100% |████████████████████████████████| 757kB 83.4MB/s ta 0:00:01  91% |█████████████████████████████▏  | 686kB 83.9MB/s eta 0:00:01
[?25hRequirement already satisfied: numpy>=1.15.0 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from imgaug==0.2.9)
Requirement already satisfied: opencv-python in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from imgaug==0.2.9)
Requirement already satisfied: matplotlib in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from imgaug==0.2.9)
Requirement already satisfied: Pillow in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from imgaug==0.2.9)
Requirement already satisfied: scikit-image>=0.11.0 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from imgaug==0.2.9)
Requirement already satisfied: six in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from imgaug==0.2.9)
Collecting Shapely (from imgaug==0.2.9)
  Downloading http://repo.myhuaweicloud.com/repository/pypi/packages/9d/18/557d4f55453fe00f59807b111cc7b39ce53594e13ada88e16738fb4ff7fb/Shapely-1.7.1-cp36-cp36m-manylinux1_x86_64.whl (1.0MB)
    100% |████████████████████████████████| 1.0MB 40.5MB/s ta 0:00:01
[?25hRequirement already satisfied: imageio in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from imgaug==0.2.9)
Requirement already satisfied: scipy in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from imgaug==0.2.9)
Requirement already satisfied: python-dateutil>=2.1 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from matplotlib->imgaug==0.2.9)
Requirement already satisfied: pytz in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from matplotlib->imgaug==0.2.9)
Requirement already satisfied: kiwisolver>=1.0.1 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from matplotlib->imgaug==0.2.9)
Requirement already satisfied: cycler>=0.10 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from matplotlib->imgaug==0.2.9)
Requirement already satisfied: pyparsing!=2.0.4,!=2.1.2,!=2.1.6,>=2.0.1 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from matplotlib->imgaug==0.2.9)
Requirement already satisfied: cloudpickle>=0.2.1 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from scikit-image>=0.11.0->imgaug==0.2.9)
Requirement already satisfied: PyWavelets>=0.4.0 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from scikit-image>=0.11.0->imgaug==0.2.9)
Requirement already satisfied: networkx>=1.8 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from scikit-image>=0.11.0->imgaug==0.2.9)
Requirement already satisfied: decorator>=4.1.0 in /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages (from networkx>=1.8->scikit-image>=0.11.0->imgaug==0.2.9)
Installing collected packages: Shapely, imgaug
  Found existing installation: imgaug 0.2.6
    Uninstalling imgaug-0.2.6:
      Successfully uninstalled imgaug-0.2.6
Successfully installed Shapely-1.7.1 imgaug-0.2.9
You are using pip version 9.0.1, however version 21.0.1 is available.
You should consider upgrading via the 'pip install --upgrade pip' command.

2.对需要的代码和数据进行下载

import os
from modelarts.session import Session
session = Session()

if session.region_name == 'cn-north-1':
    bucket_path="modelarts-labs/end2end/mask_rcnn/instance_segmentation.tar.gz"
    
elif session.region_name == 'cn-north-4':
    bucket_path="modelarts-labs-bj4/end2end/mask_rcnn/instance_segmentation.tar.gz"
else:
    print("请更换地区到北京一或北京四")

if not os.path.exists('./src/mrcnn'):
    session.download_data(bucket_path=bucket_path,
                          path='./instance_segmentation.tar.gz')
    
if os.path.exists('./instance_segmentation.tar.gz'):
    # 使用tar命令解压资源包
    os.system("tar zxf ./instance_segmentation.tar.gz")
    # 清理压缩包
    os.system("rm ./instance_segmentation.tar.gz")

Successfully download file modelarts-labs-bj4/end2end/mask_rcnn/instance_segmentation.tar.gz from OBS to local ./instance_segmentation.tar.gz

3.Mask R-CNN模型训练部分

3.1 第一步：导入相应的Python库，准备预训练模型

import sys
import random
import math
import re
import time
import numpy as np
import cv2
import matplotlib
import matplotlib.pyplot as plt

from src.mrcnn.config import Config
from src.mrcnn import utils
import src.mrcnn.model as modellib
from src.mrcnn import visualize
from src.mrcnn.model import log

%matplotlib inline 

# Directory to save logs and trained model
MODEL_DIR = "logs"

# Local path to trained weights file
COCO_MODEL_PATH = "data/mask_rcnn_coco.h5"

/home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:526: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint8 = np.dtype([("qint8", np.int8, 1)])
/home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:527: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint8 = np.dtype([("quint8", np.uint8, 1)])
/home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:528: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint16 = np.dtype([("qint16", np.int16, 1)])
/home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:529: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_quint16 = np.dtype([("quint16", np.uint16, 1)])
/home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:530: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  _np_qint32 = np.dtype([("qint32", np.int32, 1)])
/home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/framework/dtypes.py:535: FutureWarning: Passing (type, 1) or '1type' as a synonym of type is deprecated; in a future version of numpy, it will be understood as (type, (1,)) / '(1,)type'.
  np_resource = np.dtype([("resource", np.ubyte, 1)])
Using TensorFlow backend.

3.2 第二步：生成相关配置项

我们定义Config类的子类MyTrainConfig，指定相关的参数，较为关键的参数有：

• NAME： Config的唯一名称
• NUM_CLASSES： 分类的数量，COCO中共有80种物体+背景
• IMAGE_MIN_DIM和IMAGE_MAX_DIM： 图片的最大和最小尺寸，我们生成固定的128x128的图片，因此都设置为128
• TRAIN_ROIS_PER_IMAGE： 每张图片上训练的RoI个数
• STEPS_PER_EPOCH和VALIDATION_STEPS： 训练和验证时，每轮的step数量，减少step的数量可以加速训练，但是检测精度降低

class MyTrainConfig(Config):
    # 可辨识的名称
    NAME = "my_train"

    # GPU的数量和每个GPU处理的图片数量，可以根据实际情况进行调整，参考为Nvidia Tesla P100
    GPU_COUNT = 1
    IMAGES_PER_GPU = 1

    # 物体的分类个数，COCO中共有80种物体+背景
    NUM_CLASSES = 1 + 80  # background + 80 shapes

    # 图片尺寸统一处理为1024，可以根据实际情况再进一步调小
    IMAGE_MIN_DIM = 1024
    IMAGE_MAX_DIM = 1024

    # 因为我们生成的形状图片较小，这里可以使用较小的Anchor进行RoI检测
#     RPN_ANCHOR_SCALES = (8, 16, 32, 64, 128)  # anchor side in pixels

    # 每张图片上训练的RoI个数，因为我们生成的图片较小，而且每张图片上的形状较少
    #　因此可以适当调小该参数，用较少的Anchor即可覆盖大致的物体信息
    TRAIN_ROIS_PER_IMAGE = 200

    # 每轮训练的step数量
    STEPS_PER_EPOCH = 100

    # 每轮验证的step数量
    VALIDATION_STEPS = 20
config = MyTrainConfig()
config.display()

Configurations:
BACKBONE                       resnet101
BACKBONE_STRIDES               [4, 8, 16, 32, 64]
BATCH_SIZE                     1
BBOX_STD_DEV                   [0.1 0.1 0.2 0.2]
COMPUTE_BACKBONE_SHAPE         None
DETECTION_MAX_INSTANCES        100
DETECTION_MIN_CONFIDENCE       0.7
DETECTION_NMS_THRESHOLD        0.3
FPN_CLASSIF_FC_LAYERS_SIZE     1024
GPU_COUNT                      1
GRADIENT_CLIP_NORM             5.0
IMAGES_PER_GPU                 1
IMAGE_CHANNEL_COUNT            3
IMAGE_MAX_DIM                  1024
IMAGE_META_SIZE                93
IMAGE_MIN_DIM                  1024
IMAGE_MIN_SCALE                0
IMAGE_RESIZE_MODE              square
IMAGE_SHAPE                    [1024 1024    3]
LEARNING_MOMENTUM              0.9
LEARNING_RATE                  0.001
LOSS_WEIGHTS                   {'rpn_class_loss': 1.0, 'rpn_bbox_loss': 1.0, 'mrcnn_class_loss': 1.0, 'mrcnn_bbox_loss': 1.0, 'mrcnn_mask_loss': 1.0}
MASK_POOL_SIZE                 14
MASK_SHAPE                     [28, 28]
MAX_GT_INSTANCES               100
MEAN_PIXEL                     [123.7 116.8 103.9]
MINI_MASK_SHAPE                (56, 56)
NAME                           my_train
NUM_CLASSES                    81
POOL_SIZE                      7
POST_NMS_ROIS_INFERENCE        1000
POST_NMS_ROIS_TRAINING         2000
PRE_NMS_LIMIT                  6000
ROI_POSITIVE_RATIO             0.33
RPN_ANCHOR_RATIOS              [0.5, 1, 2]
RPN_ANCHOR_SCALES              (32, 64, 128, 256, 512)
RPN_ANCHOR_STRIDE              1
RPN_BBOX_STD_DEV               [0.1 0.1 0.2 0.2]
RPN_NMS_THRESHOLD              0.7
RPN_TRAIN_ANCHORS_PER_IMAGE    256
STEPS_PER_EPOCH                100
TOP_DOWN_PYRAMID_SIZE          256
TRAIN_BN                       False
TRAIN_ROIS_PER_IMAGE           200
USE_MINI_MASK                  True
USE_RPN_ROIS                   True
VALIDATION_STEPS               20
WEIGHT_DECAY                   0.0001

3.3 第三步：准备数据集

我们使用封装好的CocoDataset类，生成训练集和验证集。

from src.mrcnn.coco import CocoDataset
COCO_DIR = 'data'

# 生成训练集
dataset_train = CocoDataset()
dataset_train.load_coco(COCO_DIR, "train") #　加载训练数据集
dataset_train.prepare()

loading annotations into memory...

Done (t=0.04s)

creating index...

index created!

# 生成验证集
dataset_val = CocoDataset()
dataset_val.load_coco(COCO_DIR, "val") #　加载验证数据集
dataset_val.prepare()

loading annotations into memory...

Done (t=0.17s)

creating index...

index created!

4.创建模型

4.1 第一步：用"training"模式创建模型对象，用于形状数据集的训练

model = modellib.MaskRCNN(mode="training", config=config,
                          model_dir=MODEL_DIR)

WARNING:tensorflow:From /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.

Instructions for updating:

Colocations handled automatically by placer.


[DEBUG] <__main__.MyTrainConfig object at 0x7f9b6edc7c50>

[DEBUG] Tensor("rpn_class/concat:0", shape=(?, ?, 2), dtype=float32) Tensor("rpn_bbox_1/concat:0", shape=(?, ?, 4), dtype=float32) <tf.Variable 'anchors/Variable:0' shape=(1, 261888, 4) dtype=float32_ref>

4.2 第二步：加载预训练模型的权重

model.load_weights(COCO_MODEL_PATH, by_name=True)

接下来，我们使用预训练的模型，结合Shapes数据集，对模型进行训练

5.训练模型

Keras中的模型可以按照制定的层进行构建，在模型的train方法中，我们可以通过layers参数来指定特定的层进行训练。layers参数有以下几种预设值：

• heads:只训练head网络中的分类、mask和bbox回归
• all: 所有的layer
• 3+: 训练ResNet Stage3和后续Stage
• 4+: 训练ResNet Stage4和后续Stage
• 5+: 训练ResNet Stage5和后续Stage

此外，layers参数还支持正则表达式，按照匹配规则指定layer，可以调用model.keras_model.summary()查看各个层的名称，然后按照需要指定要训练的层。
下面的步骤对所有的layer训练1个epoch，耗时约4分钟

model.train(dataset_train, dataset_val, 
            learning_rate=config.LEARNING_RATE, 
            epochs=1, 
            layers='all')

model_savepath = 'my_mrcnn_model.h5'
model.keras_model.save_weights(model_savepath)

Starting at epoch 0. LR=0.001



Checkpoint Path: logs/my_train20210309T1458/mask_rcnn_my_train_{epoch:04d}.h5

WARNING:tensorflow:From /home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/ops/math_ops.py:3066: to_int32 (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.cast instead.
/home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/tensorflow/python/ops/gradients_impl.py:110: UserWarning: Converting sparse IndexedSlices to a dense Tensor of unknown shape. This may consume a large amount of memory.
  "Converting sparse IndexedSlices to a dense Tensor of unknown shape. "
/home/ma-user/anaconda3/envs/TensorFlow-1.13.1/lib/python3.6/site-packages/keras/engine/training_generator.py:47: UserWarning: Using a generator with `use_multiprocessing=True` and multiple workers may duplicate your data. Please consider using the`keras.utils.Sequence class.
  UserWarning('Using a generator with `use_multiprocessing=True`'
Epoch 1/1
100/100 [==============================] - 111s 1s/step - loss: 0.4283 - rpn_class_loss: 0.0090 - rpn_bbox_loss: 0.0787 - mrcnn_class_loss: 0.0627 - mrcnn_bbox_loss: 0.0758 - mrcnn_mask_loss: 0.2021 - val_loss: 0.4290 - val_rpn_class_loss: 0.0100 - val_rpn_bbox_loss: 0.1086 - val_mrcnn_class_loss: 0.0920 - val_mrcnn_bbox_loss: 0.0539 - val_mrcnn_mask_loss: 0.1645

6.使用Mask R-CNN 检测图片物体

6.1 第一步：定义InferenceConfig，并创建"Inference"模式的模型对象

class InferenceConfig(MyTrainConfig):
    GPU_COUNT = 1
    IMAGES_PER_GPU = 1

inference_config = InferenceConfig()

inference_model = modellib.MaskRCNN(mode="inference", 
                                    config=inference_config,
                                    model_dir=MODEL_DIR)

[DEBUG] <__main__.InferenceConfig object at 0x7f9681f59710>
[DEBUG] Tensor("rpn_class_1/concat:0", shape=(?, ?, 2), dtype=float32) Tensor("rpn_bbox_3/concat:0", shape=(?, ?, 4), dtype=float32) Tensor("input_anchors:0", shape=(?, ?, 4), dtype=float32)


WARNING:tensorflow:From /home/ma-user/work/case_dev/mask_rcnn/src/mrcnn/model.py:772: to_float (from tensorflow.python.ops.math_ops) is deprecated and will be removed in a future version.
Instructions for updating:
Use tf.cast instead.

将我们生成的模型权重信息加载进来

# 加载我们自己训练出的形状模型文件的权重信息
print("Loading weights from ", model_savepath)
inference_model.load_weights(model_savepath, by_name=True)

Loading weights from  my_mrcnn_model.h5

6.2 第二步：从验证数据集中随机选出一张图片进行预测，并显示结果

# 随机选出图片进行测试
image_id = random.choice(dataset_val.image_ids)
original_image, image_meta, gt_class_id, gt_bbox, gt_mask =\
    modellib.load_image_gt(dataset_val, inference_config, 
                           image_id, use_mini_mask=False)

log("original_image", original_image)
log("image_meta", image_meta)
log("gt_class_id", gt_class_id)
log("gt_bbox", gt_bbox)
log("gt_mask", gt_mask)

det_instances_savepath = 'random.det_instances.jpg'
visualize.display_instances(original_image, gt_bbox, gt_mask, gt_class_id, 
                            dataset_train.class_names, figsize=(8, 8),
                           save_path=det_instances_savepath)

original_image           shape: (1024, 1024, 3)       min:    0.00000  max:  255.00000  uint8
image_meta               shape: (93,)                 min:    0.00000  max: 1024.00000  float64
gt_class_id              shape: (17,)                 min:    1.00000  max:   74.00000  int32
gt_bbox                  shape: (17, 4)               min:    1.00000  max: 1024.00000  int32
gt_mask                  shape: (1024, 1024, 17)      min:    0.00000  max:    1.00000  bool

# 定义助手函数用于设置matplot中的子绘制区域所在的行和列
def get_ax(rows=1, cols=1, size=8):
    _, ax = plt.subplots(rows, cols, figsize=(size*cols, size*rows))
    return ax

results = inference_model.detect([original_image], verbose=1)

r = results[0]
prediction_savepath = 'random.prediction.jpg'
visualize.display_instances(original_image, r['rois'], r['masks'], r['class_ids'], 
                            dataset_val.class_names, r['scores'], ax=get_ax(),
                            save_path=prediction_savepath)

Processing 1 images
image                    shape: (1024, 1024, 3)       min:    0.00000  max:  255.00000  uint8
molded_images            shape: (1, 1024, 1024, 3)    min: -123.70000  max:  151.10000  float64
image_metas              shape: (1, 93)               min:    0.00000  max: 1024.00000  int64
anchors                  shape: (1, 261888, 4)        min:   -0.35390  max:    1.29134  float32

6.3 第三步：测试其他图片。

本目录下的data/val2014目录下有很多测试图片，修改下面代码中test_path变量右边的文件名，即可更换为不同图片，测试图片的预测效果。

test_path = './data/val2014/COCO_val2014_000000019176.jpg'

import skimage.io
image = skimage.io.imread(test_path)

results = inference_model.detect([image], verbose=1)

r = results[0]
prediction_savepath = 'self.prediction.jpg'
visualize.display_instances(image, r['rois'], r['masks'], r['class_ids'], 
                            dataset_val.class_names, r['scores'], ax=get_ax(),
                            save_path=prediction_savepath)

Processing 1 images
image                    shape: (480, 640, 3)         min:    0.00000  max:  255.00000  uint8
molded_images            shape: (1, 1024, 1024, 3)    min: -123.70000  max:  151.10000  float64
image_metas              shape: (1, 93)               min:    0.00000  max: 1024.00000  float64
anchors                  shape: (1, 261888, 4)        min:   -0.35390  max:    1.29134  float32

7.评估模型

这一步我们对自己训练出的模型进行一个简单的评估。计算模型的平均精度mAP（mean Average Precision)

# 计算VOC类型的 mAP，条件是 IoU=0.5
# 下面的示例中只选出10张图片进行评估，增加图片数量可以增加模型评估的准确性
image_ids = np.random.choice(dataset_val.image_ids, 10)
APs = []
for image_id in image_ids:
    # Load image and ground truth data
    image, image_meta, gt_class_id, gt_bbox, gt_mask =\
        modellib.load_image_gt(dataset_val, inference_config,
                               image_id, use_mini_mask=False)
    molded_images = np.expand_dims(modellib.mold_image(image, inference_config), 0)
    # Run object detection
    results = inference_model.detect([image], verbose=0)
    r = results[0]
    # Compute AP
    AP, precisions, recalls, overlaps =\
        utils.compute_ap(gt_bbox, gt_class_id, gt_mask,
                         r["rois"], r["class_ids"], r["scores"], r['masks'])
    APs.append(AP)
    
print("mAP: ", np.mean(APs))

mAP:  0.6203394930987131

本案例结束。