3 计算机视觉-阅读笔记（7）
2019-04-28 本文已影响0人
深度学习模型优化
3.6 Mask R-CNN代码简介
这里记录下，以备后续使用。
这里介绍Tensorflow和Keras实现的代码。
安装
git clone https://github.com/matterport/Mask_RCNN.git
# 或者使用作者fork的版本
git clone https://github.com/fancyerii/Mask_RCNN.git
#建议创建一个virtualenv
pip install -r requirements.txt
# 还需要安装pycocotools
# 否则会出现ImportError: No module named 'pycocotools'
# 参考 https://github.com/matterport/Mask_RCNN/issues/6
pip install "git+https://github.com/philferriere/cocoapi.git#egg=pycocotools&subdirectory=PythonAPI"
demo.ipynb
下载模型参数
COCO_MODEL_PATH = os.path.join(ROOT_DIR, "mask_rcnn_coco.h5")
# Download COCO trained weights from Releases if needed
if not os.path.exists(COCO_MODEL_PATH):
utils.download_trained_weights(COCO_MODEL_PATH)
创建模型和加载参数：
# 创建MaskRCNN对象，模式是inference
model = modellib.MaskRCNN(mode="inference", model_dir=MODEL_DIR, config=config)
# 加载模型参数
model.load_weights(COCO_MODEL_PATH, by_name=True)
读取图片并且进行分割：
# 随机加载一张图片
file_names = next(os.walk(IMAGE_DIR))[2]
image = skimage.io.imread(os.path.join(IMAGE_DIR, random.choice(file_names)))
# 进行目标检测和分割
results = model.detect([image], verbose=1)
# 显示结果
r = results[0]
visualize.display_instances(image, r['rois'], r['masks'], r['class_ids'],
class_names, r['scores'])
图1 Mask R-CNN检测结果
train_shapes.ipynb
该代码使用自己的数据进行训练。
配置
class ShapesConfig(Config):
"""用于训练shape数据集的配置
继承子基本的Config类，然后override了一些配置项。
"""
# 起个好记的名字
NAME = "shapes"
# 使用一个GPU训练，每个GPU上8个图片。因此batch大小是8 (GPUs * images/GPU).
GPU_COUNT = 1
IMAGES_PER_GPU = 8
# 分类数(需要包括背景类)
NUM_CLASSES = 1 + 3 # background + 3 shapes
# 图片为固定的128x128
IMAGE_MIN_DIM = 128
IMAGE_MAX_DIM = 128
# 因为图片比较小，所以RPN anchor也是比较小的
RPN_ANCHOR_SCALES = (8, 16, 32, 64, 128) # anchor side in pixels
# 每张图片建议的RoI数量，对于这个小图片的例子可以取比较小的值。
TRAIN_ROIS_PER_IMAGE = 32
# 每个epoch的数据量
STEPS_PER_EPOCH = 100
# 每5步验证一下。
VALIDATION_STEPS = 5
config = ShapesConfig()
config.display()
数据
- load_image
- load_mask
- image_reference
class ShapesDataset(utils.Dataset):
"""随机生成shape数据。包括三角形，正方形和圆形，以及它的位置。
这是on-th-fly的生成数据，因此不需要访问文件。
"""
def load_shapes(self, count, height, width):
"""生成图片
count: 返回的图片数量
height, width: 生成图片的height和width
"""
# 类别
self.add_class("shapes", 1, "square")
self.add_class("shapes", 2, "circle")
self.add_class("shapes", 3, "triangle")
# 注意：这里只是生成图片的specifications(说明书)，
# 具体包括性质、颜色、大小和位置等信息。
# 真正的图片是在load_image()函数里根据这些specifications
# 来on-th-fly的生成。
for i in range(count):
bg_color, shapes = self.random_image(height, width)
self.add_image("shapes", image_id=i, path=None,
width=width, height=height,
bg_color=bg_color, shapes=shapes)
def add_image(self, source, image_id, path, **kwargs):
image_info = {
"id": image_id,
"source": source,
"path": path,
}
image_info.update(kwargs)
self.image_info.append(image_info)
def random_image(self, height, width):
"""随机的生成一个specifications
它包括图片的背景演示和一些(最多4个)不同的shape的specifications。
"""
# 随机选择背景颜色
bg_color = np.array([random.randint(0, 255) for _ in range(3)])
# 随机生成一些(最多4个)shape
shapes = []
boxes = []
N = random.randint(1, 4)
for _ in range(N):
# random_shape函数随机产生一个shape(比如圆形)，它的颜色和位置
shape, color, dims = self.random_shape(height, width)
shapes.append((shape, color, dims))
# 位置是中心点和大小(正方形，圆形和等边三角形只需要一个值表示大小)
x, y, s = dims
# 根据中心点和大小计算bounding box
boxes.append([y-s, x-s, y+s, x+s])
# 使用non-max suppression去掉重叠很严重的图片
keep_ixs = utils.non_max_suppression(np.array(boxes), np.arange(N), 0.3)
shapes = [s for i, s in enumerate(shapes) if i in keep_ixs]
return bg_color, shapes
def random_shape(self, height, width):
"""随机生成一个shape的specifications，
要求这个shape在height和width的范围内。
返回一个3-tuple:
* shape名字 (square, circle, ...)
* shape的颜色：代表RGB的3-tuple
* shape的大小，一个数值
"""
# 随机选择shape的名字
shape = random.choice(["square", "circle", "triangle"])
# 随机选择颜色
color = tuple([random.randint(0, 255) for _ in range(3)])
# 随机选择中心点位置，在范围[buffer, height/widht - buffer -1]内随机选择
buffer = 20
y = random.randint(buffer, height - buffer - 1)
x = random.randint(buffer, width - buffer - 1)
# 随机的大小size
s = random.randint(buffer, height//4)
return shape, color, (x, y, s)
def load_image(self, image_id):
"""根据specs生成实际的图片
如果是实际的数据集，通常是从一个文件读取。
"""
info = self.image_info[image_id]
bg_color = np.array(info['bg_color']).reshape([1, 1, 3])
# 首先填充背景色
image = np.ones([info['height'], info['width'], 3], dtype=np.uint8)
image = image * bg_color.astype(np.uint8)
# 分别绘制每一个shape
for shape, color, dims in info['shapes']:
image = self.draw_shape(image, shape, dims, color)
return image
def draw_shape(self, image, shape, dims, color):
"""根据specs绘制shape"""
# 获取中心点x, y和size s
x, y, s = dims
if shape == 'square':
cv2.rectangle(image, (x-s, y-s), (x+s, y+s), color, -1)
elif shape == "circle":
cv2.circle(image, (x, y), s, color, -1)
elif shape == "triangle":
points = np.array([[(x, y-s),
(x-s/math.sin(math.radians(60)), y+s),
(x+s/math.sin(math.radians(60)), y+s),
]], dtype=np.int32)
cv2.fillPoly(image, points, color)
return image
def load_mask(self, image_id):
"""生成给定图片的mask
"""
info = self.image_info[image_id]
shapes = info['shapes']
count = len(shapes)
# 每个物体都有一个mask矩阵，大小是height x width
mask = np.zeros([info['height'], info['width'], count], dtype=np.uint8)
for i, (shape, _, dims) in enumerate(info['shapes']):
# 绘图函数draw_shape已经把mask绘制出来了。我们只需要传入特殊颜色值1。
mask[:, :, i:i+1] = self.draw_shape(mask[:, :, i:i+1].copy(),
shape, dims, 1)
# 处理遮挡(occlusions)
occlusion = np.logical_not(mask[:, :, -1]).astype(np.uint8)
for i in range(count-2, -1, -1):
mask[:, :, i] = mask[:, :, i] * occlusion
occlusion = np.logical_and(occlusion, np.logical_not(mask[:, :, i]))
# 类名到id
class_ids = np.array([self.class_names.index(s[0]) for s in shapes])
return mask.astype(np.bool), class_ids.astype(np.int32)
def image_reference(self, image_id):
info = self.image_info[image_id]
if info["source"] == "shapes":
return info["shapes"]
else:
super(self.__class__).image_reference(self, image_id)
# 训练集500个图片
dataset_train = ShapesDataset()
dataset_train.load_shapes(500, config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
dataset_train.prepare()
# 验证集50个图片
dataset_val = ShapesDataset()
dataset_val.load_shapes(50, config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
dataset_val.prepare()
image_ids = np.random.choice(dataset_train.image_ids, 4)
for image_id in image_ids:
image = dataset_train.load_image(image_id)
mask, class_ids = dataset_train.load_mask(image_id)
visualize.display_top_masks(image, mask, class_ids, dataset_train.class_names)
图2 随机生成的图片
model = modellib.MaskRCNN(mode="training", config=config,
model_dir=MODEL_DIR)
# 默认使用coco模型来初始化
init_with = "coco" # imagenet, coco, or last
if init_with == "imagenet":
model.load_weights(model.get_imagenet_weights(), by_name=True)
elif init_with == "coco":
# 加载COCO模型的参数，去掉全连接层(mrcnn_bbox_fc)，
# logits(mrcnn_class_logits)
# 输出的boudning box(mrcnn_bbox)和Mask(mrcnn_mask)
model.load_weights(COCO_MODEL_PATH, by_name=True,
exclude=["mrcnn_class_logits", "mrcnn_bbox_fc",
"mrcnn_bbox", "mrcnn_mask"])
elif init_with == "last":
# 加载我们最近训练的模型来初始化
model.load_weights(model.find_last(), by_name=True)
model.train(dataset_train, dataset_val,
learning_rate=config.LEARNING_RATE,
epochs=1,
layers='heads')
# 手动保存参数，这通常是不需要的，
# 因为每次epoch介绍会自动保存，所以这里是注释掉的。
# model_path = os.path.join(MODEL_DIR, "mask_rcnn_shapes.h5")
# model.keras_model.save_weights(model_path)
class InferenceConfig(ShapesConfig):
GPU_COUNT = 1
IMAGES_PER_GPU = 1
inference_config = InferenceConfig()
# 重新构建用于inference的模型
model = modellib.MaskRCNN(mode="inference",
config=inference_config,
model_dir=MODEL_DIR)
# 加载模型参数，可以手动指定也可以让它自己找最近的模型参数文件
# model_path = os.path.join(ROOT_DIR, ".h5 file name here")
model_path = model.find_last()
# 加载模型参数
print("Loading weights from ", model_path)
model.load_weights(model_path, by_name=True)
# 随机选择验证集的一张图片。
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)
visualize.display_instances(original_image, gt_bbox, gt_mask, gt_class_id,
dataset_train.class_names, figsize=(8, 8))
图3 标签结果
results = model.detect([original_image], verbose=1)
r = results[0]
visualize.display_instances(original_image, r['rois'], r['masks'], r['class_ids'],
dataset_val.class_names, r['scores'], ax=get_ax())
图4 模型预测结果
image_ids = np.random.choice(dataset_val.image_ids, 10)
APs = []
for image_id in image_ids:
# 加载图片和正确的Bounding box以及mask
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)
# 进行检测
results = model.detect([image], verbose=0)
r = results[0]
# 计算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))
# 输出0.95
inspect_data.ipynb
config = ShapesConfig()
# 我们把下面的代码注释掉
# MS COCO Dataset
#import coco
#config = coco.CocoConfig()
#COCO_DIR = "path to COCO dataset" # TODO: enter value here
# Load dataset
if config.NAME == 'shapes':
dataset = ShapesDataset()
dataset.load_shapes(500, config.IMAGE_SHAPE[0], config.IMAGE_SHAPE[1])
elif config.NAME == "coco":
dataset = coco.CocoDataset()
dataset.load_coco(COCO_DIR, "train")
# 使用dataset之前必须调用prepare()
dataset.prepare()
print("Image Count: {}".format(len(dataset.image_ids)))
print("Class Count: {}".format(dataset.num_classes))
for i, info in enumerate(dataset.class_info):
print("{:3}. {:50}".format(i, info['name']))
# 运行后的结果为：
Image Count: 500
Class Count: 4
0. BG
1. square
2. circle
3. triangle
image_ids = np.random.choice(dataset.image_ids, 4)
for image_id in image_ids:
image = dataset.load_image(image_id)
mask, class_ids = dataset.load_mask(image_id)
visualize.display_top_masks(image, mask, class_ids, dataset.class_names)
图5 显示4个样本
# 随机加载一个图片和它对应的mask.
image_id = random.choice(dataset.image_ids)
image = dataset.load_image(image_id)
mask, class_ids = dataset.load_mask(image_id)
# 计算Bounding box
bbox = utils.extract_bboxes(mask)
# 显示图片其它的统计信息
print("image_id ", image_id, dataset.image_reference(image_id))
log("image", image)
log("mask", mask)
log("class_ids", class_ids)
log("bbox", bbox)
# 显示图片
visualize.display_instances(image, bbox, mask, class_ids, dataset.class_names)
图6 显示样本
# 随机加载一个图片和它的mask
image_id = np.random.choice(dataset.image_ids, 1)[0]
image = dataset.load_image(image_id)
mask, class_ids = dataset.load_mask(image_id)
original_shape = image.shape
# 缩放图片，
image, window, scale, padding, _ = utils.resize_image(
image,
min_dim=config.IMAGE_MIN_DIM,
max_dim=config.IMAGE_MAX_DIM,
mode=config.IMAGE_RESIZE_MODE)
# 缩放图片后一定要缩放mask，否则就不一致了
mask = utils.resize_mask(mask, scale, padding)
# 计算Bounding box
bbox = utils.extract_bboxes(mask)
# 显示图片的其它统计信息
print("image_id: ", image_id, dataset.image_reference(image_id))
print("Original shape: ", original_shape)
log("image", image)
log("mask", mask)
log("class_ids", class_ids)
log("bbox", bbox)
# 显示图片
visualize.display_instances(image, bbox, mask, class_ids, dataset.class_names)
image_id = np.random.choice(dataset.image_ids, 1)[0]
image, image_meta, class_ids, bbox, mask = modellib.load_image_gt(
dataset, config, image_id, use_mini_mask=False)
log("image", image)
log("image_meta", image_meta)
log("class_ids", class_ids)
log("bbox", bbox)
log("mask", mask)
display_images([image]+[mask[:,:,i] for i in range(min(mask.shape[-1], 7))])
# 输出
image shape: (128, 128, 3) min: 4.00000 max: 241.00000 uint8
image_meta shape: (16,) min: 0.00000 max: 409.00000 int64
class_ids shape: (2,) min: 1.00000 max: 3.00000 int32
bbox shape: (2, 4) min: 14.00000 max: 128.00000 int32
mask shape: (128, 128, 2) min: 0.00000 max: 1.00000 bool
图7 显示样本
image, image_meta, class_ids, bbox, mask = modellib.load_image_gt(
dataset, config, image_id, augment=True, use_mini_mask=True)
log("mask", mask)
display_images([image]+[mask[:,:,i] for i in range(min(mask.shape[-1], 7))])
图8 mini mask 图像增强
# Generate Anchors
backbone_shapes = modellib.compute_backbone_shapes(config, config.IMAGE_SHAPE)
anchors = utils.generate_pyramid_anchors(config.RPN_ANCHOR_SCALES,
config.RPN_ANCHOR_RATIOS,
backbone_shapes,
config.BACKBONE_STRIDES,
config.RPN_ANCHOR_STRIDE)
# 输出anchor的摘要信息
num_levels = len(backbone_shapes)
anchors_per_cell = len(config.RPN_ANCHOR_RATIOS)
print("Count: ", anchors.shape[0])
print("Scales: ", config.RPN_ANCHOR_SCALES)
print("ratios: ", config.RPN_ANCHOR_RATIOS)
print("Anchors per Cell: ", anchors_per_cell)
print("Levels: ", num_levels)
anchors_per_level = []
for l in range(num_levels):
num_cells = backbone_shapes[l][0] * backbone_shapes[l][1]
anchors_per_level.append(anchors_per_cell * num_cells //
config.RPN_ANCHOR_STRIDE**2)
print("Anchors in Level {}: {}".format(l, anchors_per_level[l]))
Count: 4092
Scales: (8, 16, 32, 64, 128)
ratios: [0.5, 1, 2]
Anchors per Cell: 3
Levels: 5
Anchors in Level 0: 3072
Anchors in Level 1: 768
Anchors in Level 2: 192
Anchors in Level 3: 48
Anchors in Level 4: 12
## Visualize anchors of one cell at the center of the feature map of a specific level
# Load and draw random image
image_id = np.random.choice(dataset.image_ids, 1)[0]
image, image_meta, _, _, _ = modellib.load_image_gt(dataset, config, image_id)
fig, ax = plt.subplots(1, figsize=(10, 10))
ax.imshow(image)
levels = len(backbone_shapes)
for level in range(levels):
colors = visualize.random_colors(levels)
# Compute the index of the anchors at the center of the image
level_start = sum(anchors_per_level[:level]) # sum of anchors of previous levels
level_anchors = anchors[level_start:level_start+anchors_per_level[level]]
print("Level {}. Anchors: {:6} Feature map Shape: {}".format(level,
level_anchors.shape[0], backbone_shapes[level]))
center_cell = backbone_shapes[level] // 2
center_cell_index = (center_cell[0] * backbone_shapes[level][1] + center_cell[1])
level_center = center_cell_index * anchors_per_cell
center_anchor = anchors_per_cell * (
(center_cell[0] * backbone_shapes[level][1] / config.RPN_ANCHOR_STRIDE**2) \
+ center_cell[1] / config.RPN_ANCHOR_STRIDE)
level_center = int(center_anchor)
# Draw anchors. Brightness show the order in the array, dark to bright.
for i, rect in enumerate(level_anchors[level_center:level_center+anchors_per_cell]):
y1, x1, y2, x2 = rect
p = patches.Rectangle((x1, y1), x2-x1, y2-y1, linewidth=2, facecolor='none',
edgecolor=(i+1)*np.array(colors[level]) / anchors_per_cell)
ax.add_patch(p)
图9 Anchors
random_rois = 2000
g = modellib.data_generator(
dataset, config, shuffle=True, random_rois=random_rois,
batch_size=4,
detection_targets=True)
# Get Next Image
if random_rois:
[normalized_images, image_meta, rpn_match, rpn_bbox, gt_class_ids,
gt_boxes, gt_masks, rpn_rois, rois],
[mrcnn_class_ids, mrcnn_bbox, mrcnn_mask] = next(g)
log("rois", rois)
log("mrcnn_class_ids", mrcnn_class_ids)
log("mrcnn_bbox", mrcnn_bbox)
log("mrcnn_mask", mrcnn_mask)
else:
[normalized_images, image_meta, rpn_match, rpn_bbox, gt_boxes, gt_masks], _ =
next(g)
log("gt_class_ids", gt_class_ids)
log("gt_boxes", gt_boxes)
log("gt_masks", gt_masks)
log("rpn_match", rpn_match, )
log("rpn_bbox", rpn_bbox)
image_id = modellib.parse_image_meta(image_meta)["image_id"][0]
print("image_id: ", image_id, dataset.image_reference(image_id))
# Remove the last dim in mrcnn_class_ids. It's only added
# to satisfy Keras restriction on target shape.
mrcnn_class_ids = mrcnn_class_ids[:,:,0]
b = 0
# Restore original image (reverse normalization)
sample_image = modellib.unmold_image(normalized_images[b], config)
# Compute anchor shifts.
indices = np.where(rpn_match[b] == 1)[0]
refined_anchors = utils.apply_box_deltas(anchors[indices], rpn_bbox[b, :len(indices)]
* config.RPN_BBOX_STD_DEV)
log("anchors", anchors)
log("refined_anchors", refined_anchors)
# Get list of positive anchors
positive_anchor_ids = np.where(rpn_match[b] == 1)[0]
print("Positive anchors: {}".format(len(positive_anchor_ids)))
negative_anchor_ids = np.where(rpn_match[b] == -1)[0]
print("Negative anchors: {}".format(len(negative_anchor_ids)))
neutral_anchor_ids = np.where(rpn_match[b] == 0)[0]
print("Neutral anchors: {}".format(len(neutral_anchor_ids)))
# ROI breakdown by class
for c, n in zip(dataset.class_names, np.bincount(mrcnn_class_ids[b].flatten())):
if n:
print("{:23}: {}".format(c[:20], n))
# Show positive anchors
visualize.draw_boxes(sample_image, boxes=anchors[positive_anchor_ids],
refined_boxes=refined_anchors)
anchors shape: (4092, 4) min: -90.50967 max: 154.50967 float64
refined_anchors shape: (3, 4) min: 6.00000 max: 128.00000 float32
Positive anchors: 3
Negative anchors: 253
Neutral anchors: 3836
BG : 22
square : 1
circle : 9
图10 正样本anchor
图11 负样本anchor
图12 无用的anchor
