mmdetection训练自己的数据，用网络deformable_detr做示例，先用labelme标注转为coco格式，训练后测试并分析
目录
1.标注labelme
2.将labelme标注的数据转为coco格式
直接上代码：
coco格式如下：
3.mmdetection训练自己的数据，用网络deformable_detr做示例
（0）先生成整体配置文件，在一个配置文件中修改比较方便，方法参考：open-mmlab. mmclassification安装并使用自己数据集windows下_黛玛日孜的博客-CSDN博客
（1）#mmdet/datasets/coco.py中将类CocoDataset中的内容改成自己的，只改类别和颜色表示
（2）#mmdet/core/evaluation/classnames.py中将函数coco_classes中的内容改成自己的
（3）展示整体配置文件。配置文件configs中的类别数量改成自己的，并修改数据路径，
（4）遇到其他问题，可以去官网问答区搜索，如下
4.模型训练train.py（选择可视化标注文件browse_dataset.py）
5.DEMO演示image_demo.py
6.模型测试test.py
7.可视化分析模块confusion_matrix.py、analyze_results.py、analyze_logs.py等
其他可在官网查看
8.参数量、计算量
1.标注labelme
安装labelme，
1.使用Annoconda创建虚拟环境
conda create -n labelme python=3.6
activate labelme
2.安装相关的包
pip install pyqt
pip install labelme
3.启动labelme 并标注：
在labelme的虚拟环境中键入labelme就会启动labelme可视化标注软件
安装好labelme后直接在cmd命名窗口输入labelme打开。
open打开单张图片，opendir打开文件夹。
右键，选择矩形框标注。
左上角到右下角标注，确定框的名字，可以标注多个框。保存
，最好不改名字。
得到如下文件格式：
2.将labelme标注的数据转为coco格式
直接上代码：
D:\Code\mmdetection-master\mmdet\data\json2coco.py
重点修改类别和输入输出文件路径，以及测试集比例：
classname_to_id = {
"mask": 0, #改成自己的类别
"person": 1
}
labelme_path = "./labelme-data/maskdataset"
saved_coco_path = "./labelme-data/coco-format"
train_path, val_path = train_test_split(json_list_path, test_size=0.1, train_size=0.9)
代码如下：
#D:\Code\mmdetection-master\mmdet\data\json2coco.py
import os
import json
import numpy as np
import glob
import shutil
import cv2
from sklearn.model_selection import train_test_split
np.random.seed(41)
classname_to_id = {
"mask": 0, #改成自己的类别
"person": 1
}
class Lableme2CoCo:
def __init__(self):
self.images = []
self.annotations = []
self.categories = []
self.img_id = 0
self.ann_id = 0
def save_coco_json(self, instance, save_path):
json.dump(instance, open(save_path, 'w', encoding='utf-8'), ensure_ascii=False, indent=1) # indent=2 更加美观显示
# 由json文件构建COCO
def to_coco(self, json_path_list):
self._init_categories()
for json_path in json_path_list:
obj = self.read_jsonfile(json_path)
self.images.append(self._image(obj, json_path))
shapes = obj['shapes']
for shape in shapes:
annotation = self._annotation(shape)
self.annotations.append(annotation)
self.ann_id += 1
self.img_id += 1
instance = {}
instance['info'] = 'spytensor created'
instance['license'] = ['license']
instance['images'] = self.images
instance['annotations'] = self.annotations
instance['categories'] = self.categories
return instance
# 构建类别
def _init_categories(self):
for k, v in classname_to_id.items():
category = {}
category['id'] = v
category['name'] = k
self.categories.append(category)
# 构建COCO的image字段
def _image(self, obj, path):
image = {}
from labelme import utils
img_x = utils.img_b64_to_arr(obj['imageData'])
h, w = img_x.shape[:-1]
image['height'] = h
image['width'] = w
image['id'] = self.img_id
image['file_name'] = os.path.basename(path).replace(".json", ".jpg")
return image
# 构建COCO的annotation字段
def _annotation(self, shape):
# print('shape', shape)
label = shape['label']
points = shape['points']
annotation = {}
annotation['id'] = self.ann_id
annotation['image_id'] = self.img_id
annotation['category_id'] = int(classname_to_id[label])
annotation['segmentation'] = [np.asarray(points).flatten().tolist()]
annotation['bbox'] = self._get_box(points)
annotation['iscrowd'] = 0
annotation['area'] = 1.0
return annotation
# 读取json文件，返回一个json对象
def read_jsonfile(self, path):
with open(path, "r", encoding='utf-8') as f:
return json.load(f)
# COCO的格式： [x1,y1,w,h] 对应COCO的bbox格式
def _get_box(self, points):
min_x = min_y = np.inf
max_x = max_y = 0
for x, y in points:
min_x = min(min_x, x)
min_y = min(min_y, y)
max_x = max(max_x, x)
max_y = max(max_y, y)
return [min_x, min_y, max_x - min_x, max_y - min_y]
#训练过程中，如果遇到Index put requires the source and destination dtypes match, got Long for the destination and Int for the source
#参考：https://github.com/open-mmlab/mmdetection/issues/6706
if __name__ == '__main__':
labelme_path = "./labelme-data/maskdataset"
saved_coco_path = "./labelme-data/coco-format"
print('reading...')
# 创建文件
if not os.path.exists("%scoco/annotations/" % saved_coco_path):
os.makedirs("%scoco/annotations/" % saved_coco_path)
if not os.path.exists("%scoco/images/train2017/" % saved_coco_path):
os.makedirs("%scoco/images/train2017" % saved_coco_path)
if not os.path.exists("%scoco/images/val2017/" % saved_coco_path):
os.makedirs("%scoco/images/val2017" % saved_coco_path)
# 获取images目录下所有的joson文件列表
print(labelme_path + "/*.json")
json_list_path = glob.glob(labelme_path + "/*.json")
print('json_list_path: ', len(json_list_path))
# 数据划分,这里没有区分val2017和tran2017目录，所有图片都放在images目录下
train_path, val_path = train_test_split(json_list_path, test_size=0.1, train_size=0.9)
print("train_n:", len(train_path), 'val_n:', len(val_path))
# 把训练集转化为COCO的json格式
l2c_train = Lableme2CoCo()
train_instance = l2c_train.to_coco(train_path)
l2c_train.save_coco_json(train_instance, '%scoco/annotations/instances_train2017.json' % saved_coco_path)
for file in train_path:
# shutil.copy(file.replace("json", "jpg"), "%scoco/images/train2017/" % saved_coco_path)
img_name = file.replace('json', 'jpg')
temp_img = cv2.imread(img_name)
try:
cv2.imwrite("{}coco/images/train2017/{}".format(saved_coco_path, img_name.split('\\')[-1].replace('png', 'jpg')), temp_img)
except Exception as e:
print(e)
print('Wrong Image:', img_name )
continue
print(img_name + '-->', img_name.replace('png', 'jpg'))
for file in val_path:
# shutil.copy(file.replace("json", "jpg"), "%scoco/images/val2017/" % saved_coco_path)
img_name = file.replace('json', 'jpg')
temp_img = cv2.imread(img_name)
try:
cv2.imwrite("{}coco/images/val2017/{}".format(saved_coco_path, img_name.split('\\')[-1].replace('png', 'jpg')), temp_img)
except Exception as e:
print(e)
print('Wrong Image:', img_name)
continue
print(img_name + '-->', img_name.replace('png', 'jpg'))
# 把验证集转化为COCO的json格式
l2c_val = Lableme2CoCo()
val_instance = l2c_val.to_coco(val_path)
l2c_val.save_coco_json(val_instance, '%scoco/annotations/instances_val2017.json' % saved_coco_path)
coco格式如下：
3.mmdetection训练自己的数据，用网络deformable_detr做示例
deformable_detr网络：MMCV需要要1.4.2。特点是很吃现存，训练速度慢。
（0）先生成整体配置文件，在一个配置文件中修改比较方便，方法参考：open-mmlab. mmclassification安装并使用自己数据集windows下_黛玛日孜的博客-CSDN博客
（1）#mmdet/datasets/coco.py中将类CocoDataset中的内容改成自己的，只改类别和颜色表示
CLASSES = ('mask', 'person')
PALETTE = [(220, 20, 60), (119, 11, 32)]
（2）#mmdet/core/evaluation/classnames.py中将函数coco_classes中的内容改成自己的
return [
'mask', 'person']
（3）展示整体配置文件。配置文件configs中的类别数量改成自己的，并修改数据路径，
如：D:\Code\mmdetection-master\configs\deformable_detr\my_deformable_detr_r50_16x2_50e_coco.py中
#num_classes改成自己的
#D:\Code\mmdetection-master\configs\deformable_detr\my_deformable_detr_r50_16x2_50e_coco.py
dataset_type = 'CocoDataset' #不读，走后面设置的绝对路径
data_root = 'data/coco/' #不读，走后面设置的绝对路径
#mmdet/core/evaluation/classnames.py中将coco_classes中的内容改成自己的
#mmdet/datasets/coco.py中将cocodatasets中的内容改成自己的
#num_classes改成自己的
img_norm_cfg = dict(
mean=[123.675, 116.28, 103.53], std=[58.395, 57.12, 57.375], to_rgb=True)
train_pipeline = [
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations', with_bbox=True),
dict(type='RandomFlip', flip_ratio=0.5),
dict(
type='AutoAugment',#自动数据增强，从以下策略policies中随机选择一个。
policies=[[{#随机带走一个小朋友。
'type':'Resize',
'img_scale': [(480, 1333), (512, 1333), (544, 1333), (576, 1333),
(608, 1333), (640, 1333), (672, 1333), (704, 1333),
(736, 1333), (768, 1333), (800, 1333)],
'multiscale_mode':'value',#多尺度训练，在上面随机找一个
'keep_ratio':True#True的时候以h和w中比例差异小的为基准倍数，对h和w按照相同比例resize（保持原有长宽比）
}], #False时直接按照上面大小resize
[{
'type': 'Resize',
'img_scale': [(400, 4200), (500, 4200), (600, 4200)],
'multiscale_mode': 'value',
'keep_ratio': True
}, {
'type': 'RandomCrop',
'crop_type': 'absolute_range',
'crop_size': (384, 600),
'allow_negative_crop': True
}, {
'type':
'Resize',
'img_scale': [(480, 1333), (512, 1333), (544, 1333),
(576, 1333), (608, 1333), (640, 1333),
(672, 1333), (704, 1333), (736, 1333),
(768, 1333), (800, 1333)],
'multiscale_mode':
'value',
'override':#无含义，使不报错
True,
'keep_ratio':
True
}]]),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='Pad', size_divisor=1),
dict(type='DefaultFormatBundle'),
dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
]
test_pipeline = [
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(1333, 800), #选训练时多尺度的中间数比较好
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='RandomFlip'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='Pad', size_divisor=1),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img'])
])
]
data = dict(
samples_per_gpu=1, #8G以下的显存选1
workers_per_gpu=1,
train=dict(
type='CocoDataset',
ann_file='D:\\Code\\mmdetection-master\\mmdet\\data\\labelme-data\\coco-formatcoco\\annotations\\instances_train2017.json',#windows最好用绝对路径，并且是双斜杠\\
img_prefix='D:\\Code\\mmdetection-master\\mmdet\\data\\labelme-data\\coco-formatcoco\\images\\train2017',
pipeline=[
dict(type='LoadImageFromFile'),
dict(type='LoadAnnotations', with_bbox=True),
dict(type='RandomFlip', flip_ratio=0.5),
dict(
type='AutoAugment',
policies=[[{
'type':
'Resize',
'img_scale': [(480, 1333), (512, 1333), (544, 1333),
(576, 1333), (608, 1333), (640, 1333),
(672, 1333), (704, 1333), (736, 1333),
(768, 1333), (800, 1333)],
'multiscale_mode':
'value',
'keep_ratio':
True
}],
[{
'type': 'Resize',
'img_scale': [(400, 4200), (500, 4200),
(600, 4200)],
'multiscale_mode': 'value',
'keep_ratio': True
}, {
'type': 'RandomCrop',
'crop_type': 'absolute_range',
'crop_size': (384, 600),
'allow_negative_crop': True
}, {
'type':
'Resize',
'img_scale': [(480, 1333), (512, 1333),
(544, 1333), (576, 1333),
(608, 1333), (640, 1333),
(672, 1333), (704, 1333),
(736, 1333), (768, 1333),
(800, 1333)],
'multiscale_mode':
'value',
'override':
True,
'keep_ratio':
True
}]]),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='Pad', size_divisor=1),
dict(type='DefaultFormatBundle'),
dict(type='Collect', keys=['img', 'gt_bboxes', 'gt_labels'])
],
filter_empty_gt=False),
val=dict(
type='CocoDataset',
ann_file='D:\\Code\\mmdetection-master\\mmdet\\data\\labelme-data\\coco-formatcoco\\annotations\\instances_val2017.json',
img_prefix='D:\\Code\\mmdetection-master\\mmdet\\data\\labelme-data\\coco-formatcoco\\images\\val2017',
pipeline=[
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(1333, 800),
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='RandomFlip'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='Pad', size_divisor=1),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img'])
])
]),
test=dict(
type='CocoDataset',
ann_file='D:\\Code\\mmdetection-master\\mmdet\\data\\labelme-data\\coco-formatcoco\\annotations\\instances_val2017.json',
img_prefix='D:\\Code\\mmdetection-master\\mmdet\\data\\labelme-data\\coco-formatcoco\\images\\val2017',
pipeline=[
dict(type='LoadImageFromFile'),
dict(
type='MultiScaleFlipAug',
img_scale=(1333, 800),
flip=False,
transforms=[
dict(type='Resize', keep_ratio=True),
dict(type='RandomFlip'),
dict(
type='Normalize',
mean=[123.675, 116.28, 103.53],
std=[58.395, 57.12, 57.375],
to_rgb=True),
dict(type='Pad', size_divisor=1),
dict(type='ImageToTensor', keys=['img']),
dict(type='Collect', keys=['img'])
])
]))
evaluation = dict(interval=10, metric='bbox') #每多少epoch评估
checkpoint_config = dict(interval=50) #每多少epoch保存模型
log_config = dict(interval=10, hooks=[dict(type='TextLoggerHook')])#每多少epoch打印信息
custom_hooks = [dict(type='NumClassCheckHook')]
dist_params = dict(backend='nccl')
log_level = 'INFO'
load_from = './work_dirs/deformable_detr_r50_16x2_50e_coco/deformable_detr_r50_16x2_50e_coco_20210419_220030-a12b9512.pth'
#https://github.com/open-mmlab/mmdetection/tree/master/configs/deformable_detr
resume_from = None
workflow = [('train', 1)]
opencv_num_threads = 0
mp_start_method = 'fork'
model = dict(
type='DeformableDETR',
backbone=dict(
type='ResNet',
depth=50,
num_stages=4,
out_indices=(1, 2, 3),
frozen_stages=1,
norm_cfg=dict(type='BN', requires_grad=False),
norm_eval=True,
style='pytorch',
init_cfg=dict(type='Pretrained', checkpoint='torchvision://resnet50')),
neck=dict(
type='ChannelMapper', #
in_channels=[512, 1024, 2048],
kernel_size=1,
out_channels=256,
act_cfg=None,
norm_cfg=dict(type='GN', num_groups=32),
num_outs=4),
bbox_head=dict(
type='DeformableDETRHead',
num_query=300,
num_classes=2,
in_channels=2048,
sync_cls_avg_factor=True,
as_two_stage=False,
transformer=dict(
type='DeformableDetrTransformer',
encoder=dict(
type='DetrTransformerEncoder',
num_layers=6,
transformerlayers=dict(
type='BaseTransformerLayer',
attn_cfgs=dict(
type='MultiScaleDeformableAttention', embed_dims=256),
feedforward_channels=1024,
ffn_dropout=0.1,
operation_order=('self_attn', 'norm', 'ffn', 'norm'))),
decoder=dict(
type='DeformableDetrTransformerDecoder',
num_layers=6,
return_intermediate=True,
transformerlayers=dict(
type='DetrTransformerDecoderLayer',
attn_cfgs=[
dict(
type='MultiheadAttention',
embed_dims=256,
num_heads=8,
dropout=0.1),
dict(
type='MultiScaleDeformableAttention',
embed_dims=256)
],
feedforward_channels=1024,
ffn_dropout=0.1,
operation_order=('self_attn', 'norm', 'cross_attn', 'norm',
'ffn', 'norm')))),
positional_encoding=dict(
type='SinePositionalEncoding',
num_feats=128,
normalize=True,
offset=-0.5),
loss_cls=dict(
type='FocalLoss',
use_sigmoid=True,
gamma=2.0,
alpha=0.25,
loss_weight=2.0),
loss_bbox=dict(type='L1Loss', loss_weight=5.0),
loss_iou=dict(type='GIoULoss', loss_weight=2.0)),
train_cfg=dict(
assigner=dict(
type='HungarianAssigner',
cls_cost=dict(type='FocalLossCost', weight=2.0),
reg_cost=dict(type='BBoxL1Cost', weight=5.0, box_format='xywh'),
iou_cost=dict(type='IoUCost', iou_mode='giou', weight=2.0))),
test_cfg=dict(max_per_img=100))
optimizer = dict(
type='AdamW',
lr=0.0002,
weight_decay=0.0001,
paramwise_cfg=dict(
custom_keys=dict(
backbone=dict(lr_mult=0.1),
sampling_offsets=dict(lr_mult=0.1),
reference_points=dict(lr_mult=0.1))))
optimizer_config = dict(grad_clip=dict(max_norm=0.1, norm_type=2))
lr_config = dict(policy='step', step=[40])
runner = dict(type='EpochBasedRunner', max_epochs=50)
work_dir = './work_dirs/deformable_detr_r50_16x2_50e_coco'
auto_resume = False
gpu_ids = [0]
（4）遇到其他问题，可以去官网问答区搜索，如下
4.模型训练train.py（选择可视化标注文件browse_dataset.py）
可选择先检查一下数据是否标注正确，可视化标注文件D:\Code\mmdetection-master\tools\misc\browse_dataset.py其参数如下
D:\\Code\\mmdetection-master\\configs\\deformable_detr\\my_deformable_detr_r50_16x2_50e_coco.py
--output-dir D:\\Code\\mmdetection-master\\mmdet\\data\\labelme-data\\coco-formatcoco\\sinpoec-label ##保存路径在D:\Code\mmdetection-master\tools\train.py用以下配置文件作为参数训练。记得用预训练模型。
D:\Code\mmdetection-master\configs\deformable_detr\my_deformable_detr_r50_16x2_50e_coco.py5.DEMO演示image_demo.py
D:\Code\mmdetection-master\demo\image_demo.py
参数如下：
D:\\Code\\mmdetection-master\\mmdet\\data\\labelme-data\\coco-formatcoco\\images\\val2017\\19.jpg
D:\\Code\\mmdetection-master\\configs\\deformable_detr\\my_deformable_detr_r50_16x2_50e_coco.py
D:\\Code\\mmdetection-master\\tools\\work_dirs\\deformable_detr_r50_16x2_50e_coco\\latest.pth6.模型测试test.py
D:\Code\mmdetection-master\tools\test.py
参数如下：
D:\\Code\\mmdetection-master\\configs\\deformable_detr\\my_deformable_detr_r50_16x2_50e_coco.py
D:\\Code\\mmdetection-master\\tools\\work_dirs\\deformable_detr_r50_16x2_50e_coco\\latest.pth
--eval bbox ##根据数据格式传入参数，"bbox",'' "segm", "proposal" for COCO, and "mAP", "recall" for PASCAL VOC'
--out ./work_dirs/deformable_detr_r50_16x2_50e_coco/test.pkl ##做数据分析要用
--show7.可视化分析模块confusion_matrix.py、analyze_results.py、analyze_logs.py等
D:\Code\mmdetection-master\tools\analysis_tools\analyze_logs.py
D:\Code\mmdetection-master\tools\analysis_tools\analyze_results.py
D:\Code\mmdetection-master\tools\analysis_tools\confusion_matrix.py
用confusion_matrix.py示例，其可做多分类，参数如下：
D:\\Code\\mmdetection-master\\configs\\deformable_detr\\my_deformable_detr_r50_16x2_50e_coco.py
../work_dirs/deformable_detr_r50_16x2_50e_coco/test.pkl
../work_dirs/deformable_detr_r50_16x2_50e_coco/ ##混淆矩阵保存地址
--show其他可在官网查看
8.参数量、计算量
D:\Code\mmdetection-master\tools\analysis_tools\get_flops.py
参数如下：
D:\\Code\\mmdetection-master\\configs\\deformable_detr\\my_deformable_detr_r50_16x2_50e_coco.py
--shape 1280 800