华东师范大学计算机与软件工程考研 [复制链接]

白癜风用什么药能治好 http://baidianfeng.39.net/a_ht/131207/4304149.html
本周关键词：ORC算法库、Python高性能CPU、SDV

pytorchOCR基于pytorch的ocr算法库

已完成模型/p>

DBnet

PSEnet

PANnet

SASTnet

CRNN

检测模型效果：训练只在ICDAR文本检测公开数据集上

模型

骨干网络

precision

recall

Hmean

DB

ResNet50_7*7

85.88%

79.10%

82.35%

DB

ResNet50_3*3

86.51%

80.59%

83.44%

DB

MobileNetV3

82.89%

75.83%

79.20%

SAST

ResNet50_7*7

85.72%

78.38%

81.89%

SAST

ResNet50_3*3

86.67%

76.74%

81.40%

PSE

ResNet50_7*7

84.10%

80.01%

82.01%

PSE

ResNet50_3*3

82.56%

78.91%

80.69%

PAN

ResNet18_7*7

81.80%

77.08%

79.37%

PAN

ResNet18_3*3

83.78%

75.15%

79.23%

模型压缩剪枝效果：

这里使用mobilev3作为backbone，在icdar上测试结果，未压缩模型初始大小为2.4M.

1.对backbone进行压缩

模型

prunedmethod

ratio

modelsize(M)

precision

recall

Hmean

DB

no

0

2.4

84.04%

75.34%

79.46%

DB

backbone

0.5

1.9

83.74%

73.18%

78.10%

DB

backbone

0.6

1.58

84.46%

69.90%

76.50%

2.对整个模型进行压缩

模型

prunedmethod

ratio

modelsize(M)

precision

recall

Hmean

DB

no

0

2.4

85.70%

74.77%

79.86%

DB

total

0.6

1.42

82.97%

75.10%

78.84%

DB

total

0.65

1.15

85.14%

72.84%

78.51%

模型蒸馏：

模型

teacher

student

modelsize(M)

precision

recall

Hmean

improve(%)

DB

no

mobilev3

2.4

85.70%

74.77%

79.86%

-

DB

resnet50

mobilev3

2.4

86.37%

77.22%

81.54%

1.68

DB

no

mobilev3

1.42

82.97%

75.10%

78.84%

-

DB

resnet50

mobilev3

1.42

85.88%

76.16%

80.73%

1.89

DB

no

mobilev3

1.15

85.14%

72.84%

78.51%

-

DB

resnet50

mobilev3

1.15

85.60%

74.72%

79.79%

1.28

项目