Mask R-CNN ubuntu16.04最全部署教程

ICCV2017最佳论文Mask R-CNN的代码复现 2018.10.29 by刘泽豪 11.28 已检查
0.基于Python3，Keras，TensorFlow。
Python 3.4+
TensorFlow 1.3+
Keras 2.0.8+
Jupyter Notebook
Numpy, skimage, scipy

anaconda搭建环境,具体参照别的教程
0.pip换源
gedit ~/.pip/pip.conf
换这个 https://pypi.tuna.tsinghua.edu.cn/simple/
在mask-rcnn下载目录下安装 pip3 install -r requirements.txt

1.运行ballon气球识别的例子,这个例子最简单,用来验证代码没有问题
1.1 下载代码
git clone https://github.com/matterport/Mask_RCNN

1.2 下载.h5权重文件mask_rcnn_balloon.h5和测试图片balloon_dataset.zip
下载地址为:https://github.com/matterport/Mask_RCNN/releases

1.3将mask_rcnn_balloon.h5放置在Mask_RCNN目录下,将balloon_dataset.zip 解压到Mask_RCNN/datasets/balloon/ 注意这时候Mask_RCNN/datasets/balloon/ 目录下应该有解压出来的 train和 val两个文件夹

1.4验证
在Mask_RCNN目录下输入 jupyter notebook,进入Mask_RCNN/samples/blloon 运行inspect_balloon_data.ipynb
运行jupyter第一个cell应该出现Using TensorFlow backend.否则说明tensorflow或者Keras安装有问题
之后每个cell正常运行进入第二个例子

2.运行demo.ipynb
2.1下载coco
git clone https://github.com/waleedka/coco

2.2下载源文件后打开 coco/PythonAPI ，并在此目录下打开终端，运行 make 。
注意:如果终端提示 Mask-RCNN 环境中缺少 Cython ，则重新安装后运行 make。

2.3将生成的 pycocotools 文件夹复制到 Mask-RCNN 的源文件中即可
注意:如果第一次生成失败,最好直接把整个coco重新删了重来,第二次生成不会覆盖第一次的

2.4修改demo.ipnb文件
在第一个cell,import coco 下面加入from mrcnn.config import Config
验证:出现
Configurationsweixin
We’ll be using a model trained on the MS-COCO dataset. The configurations of this model are in the CocoConfig class in coco.py.

For inferencing, modify the configurations a bit to fit the task. To do so, sub-class the CocoConfig class and override the attributes you need to change.
说明环境没问题

重写第二个cell为
class CocoConfig(Config):
“”“Configuration for training on MS COCO.
Derives from the base Config class and overrides values specific
to the COCO dataset.
“””
# Give the configuration a recognizable name
NAME = “coco”

# We use a GPU with 12GB memory, which can fit two images.
# Adjust down if you use a smaller GPU.
IMAGES_PER_GPU = 2

# Uncomment to train on 8 GPUs (default is 1)
# GPU_COUNT = 8

# Number of classes (including background)
#NUM_CLASSES = 37  # COCO has 80 classes
NUM_CLASSES = 81

#class InferenceConfig(coco.CocoConfig):
class InferenceConfig(CocoConfig):
# Set batch size to 1 since we’ll be running inference on
# one image at a time. Batch size = GPU_COUNT * IMAGES_PER_GPU
GPU_COUNT = 1
IMAGES_PER_GPU = 1
#MAX_GT_INSTANCES = 100
#TRAIN_ROIS_PER_IMAGE = 50
#BACKBONE = “resnet50” #not working at all!
#RPN_ANCHOR_STRIDE = 2
POST_NMS_ROIS_TRAINING = 1000
POST_NMS_ROIS_INFERENCE = 500
IMAGE_MIN_DIM = 400 #really much faster but bad results
IMAGE_MAX_DIM = 512
#DETECTION_MAX_INSTANCES = 50 #a little faster but some instances not recognized

config = InferenceConfig()
config.display()

否侧会报如下错:

AttributeError Traceback (most recent call last)
in ()
19
20 #class InferenceConfig(coco.CocoConfig):
—> 21 class InferenceConfig(coco.CocoConfig):
22 # Set batch size to 1 since we’ll be running inference on
23 # one image at a time. Batch size = GPU_COUNT * IMAGES_PER_GPU

AttributeError: module ‘coco’ has no attribute ‘CocoConfig’
感觉这里是一个BUG,自己瞎改的(可能有更好的办法),被坑了好久

3.尝试运行coco并且训练
3.1下载coco2014年的数据集，解压到Mask_RCNN/samples/coco/dataset 目录下
3.2下载 mask_rcnn_coco.h5 放到Mask_RCNN目录下 下载地址https://github.com/matterport/Mask_RCNN/releases
3.3修改COCO_DIR = “dataset” dataset为coco的存放地址
可能的错误：
1.'matplotlib.cook’has no attritube这个错误需要卸载matplotlib重装即可