地面文字检测语义分割
<h3>地面文字检测语义分割</h3>
<p>欢迎使用ShowDoc！</p>
<h4>文字检测</h4>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/7f5ad0048a33c111c07f0202dede4839" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/0c6427ceb10304b85211663c91925393" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/6bcca064233e9d4dba3b1e508f475e2d" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/01b735187538e5c15e9dc100b615a4a5" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/e8932c2cbd54f70bb56dafd220810fa0" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/2e56cc46703257fe4c9120bc50d5b646" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/263133f21a44d67eb73daa41cbd7c718" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/b9798ca88466dec0b3b87b0cbafeccd5" alt="" /></p>
<h3>01 02 数据源</h3>
<p>val_img : 463 张
train_img :1852 张</p>
<p>实例分割
框架 mmdetection
模型 cascade mask rcnn 101
mIOU： 0.8453527607978739</p>
<p>语义分割
cabinet
pytorch 1.7
notebook
0.874915963563809</p>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/1abeb81486b7c91bed4753f7bbd81566" alt="" /></p>
<p>添加样本03 04
训练集： 添加03、04 样本，从原先的1850张训练图 增加到了现在的4210张训练图
测试集 从原先463张测试图 增加到了1054张测试图</p>
<p>测试结果
在新的测试数据集上，
优化前检测模型 map=80.47%</p>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/fbddbadd1f5108c3c3f944588006836a" alt="" /></p>
<p>优化后检测模型 map=91.25% （第12次迭代）
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/e5bc3cd1d0ef5a87c35725bed24cb623" alt="" /></p>
<h3>cascade_mask_rcnn_r101(优化前) 与 cascade_mask_rcnn_r101(优化后)识别率差距原因</h3>
<p>1、相比优化前前数据集，优化后数据量到4000+，训练样本提升了一倍
2、01 、02 样本数据集整体有较高的识别度，画质清晰、03、04样本大部分偏暗，画质模糊</p>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/6d664fafecdb9d16eab826c6a6ad6be2" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/75dc630c24d1063270112b2a90f6c751" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/a2bd1edea4739c8d468ebd0bc0b5f7ee" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/877d0c890f2ba65add29f96809a962e0" alt="" /></p>
<h5>文字识别</h5>
<h4>效果图</h4>
<h1>—————————————————————</h1>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/6bd4474b1aab7dedc4f7b457da00d3bd" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/efc9ccc67f1939735e9364e139a8800e" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/6c31c2e48ec6cc3faf3f2cdf0169dde9" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/630d47d32061c303534af9e12fddcb96" alt="" /></p>
<h1>—————————————————————</h1>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/5012c59e0ee422cdb02521b547f5d574" alt="" />
<img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/802a806836a44ab23dcfbb9a67a77e6d" alt="" /></p>
<h1>—————————————————————</h1>
<h3>测试样例</h3>
<h4>60</h4>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/5caf8f01ed2eb0ad9e71b9e5b31d1343" alt="" /></p>
<h4>车</h4>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/28d383f1d8bde4e74aa6e563cc67045c" alt="" /></p>
<h4>小</h4>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/199fa7648df02481bcbbf657c56f43d5" alt="" /></p>
<h4>安</h4>
<p><img src="https://www.showdoc.com.cn/server/api/attachment/visitfile/sign/76076ca525a9ff2a1d4ddd3dcdfdd7db" alt="" /></p>
<p>识别率计算方式</p>
<h3>新测试数据集上测试</h3>
<h3>优化前文字识别模型</h3>
<h4>召回率</h4>
<p>recall_dicts : {‘超’: 0.0, ‘外’: 0.0, ‘应’: 0.0, ‘线’: 0.0, ‘型’: 0.03333333333333333, ‘19’: 0.0, ‘交’: 0.08, ‘G1’: 0.0, ‘7’: 0.0, ‘中’: 0.0, ‘门’: 0.0, ‘行’: 0.9921996879875195, ‘开’: 0.6, ‘停’: 0.0, ‘1Km’: 0.0, ‘兴’: 0.0, ‘小’: 0.959731543624161, ‘沈’: 0.0, ‘邢’: 0.0, ‘沧’: 0.0, ‘高’: 0.6666666666666666, ‘员’: 0.0, ‘乘’: 0.0, ‘东’: 0.0, ‘向’: 0.0, ‘街’: 0.0, ‘海’: 0.0, ‘丽’: 0.0, ‘夏’: 0.0, ‘延’: 0.0, ‘带’: 0.0, ‘G7’: 0.0, ‘京’: 0.5384615384615384, ‘二’: 0.0, ‘水’: 0.0, ‘靠’: 0.0, ‘架’: 0.25, ‘慢’: 0.0, ‘10’: 0.0, ‘藏’: 0.0, ‘速’: 0.14814814814814814, ‘聊’: 0.0, ‘目’: 1.0, ‘快’: 0.0, ‘16’: 0.0, ‘90’: 0.32558139534883723, ‘混’: 0.0, ‘道’: 0.9919678714859438, ‘机’: 0.0, ‘Km’: 0.0, ‘清’: 0.0, ‘S5’: 0.0, ‘内’: 0.25, ‘方’: 0.0, ‘40’: 0.0, ‘大’: 0.1320754716981132, ‘60’: 0.7098445595854922, ‘路’: 0.13636363636363635, ‘出’: 0.0, ‘西’: 0.7, ‘桥’: 0.125, ‘新’: 0.0, ‘场’: 0.0, ‘闵’: 0.0, ‘120’: 0.08695652173913043, ‘用’: 0.0, ‘天’: 0.5, ‘承’: 0.0, ‘城’: 0.0, ‘80’: 0.9072164948453608, ‘货’: 0.0, ‘山’: 0.0, ‘上’: 0.0, ‘厦’: 0.0, ‘辆’: 0.0, ‘50’: 0.5, ‘馆’: 0.0, ‘专’: 0.0, ‘口’: 0.0, ‘站’: 0.0, ‘青’: 0.0, ‘良’: 0.0, ‘区’: 0.0, ‘直’: 0.0, ‘嘉’: 0.0, ‘70’: 0.8571428571428571, ‘G4’: 0.0, ‘车’: 0.9873563218390805, ‘津’: 0.2857142857142857, ‘一’: 0.125, ‘环’: 0.0, ‘客’: 0.9824561403508771, ‘北’: 0.0, ‘右’: 0.0, ‘潘’: 0.0, ‘淮’: 0.0, ‘陶’: 0.0, ‘昆’: 0.0, ‘银’: 0.0, ‘急’: 0.0, ‘2Km’: 0.14285714285714285, ‘安’: 0.5, ‘100’: 0.9523809523809523, ‘漳’: 0.0, ‘州’: 0.0, ‘都’: 0.0, ‘南’: 0.625, ‘浦’: 0.14285714285714285, ‘临’: 0.0, ‘多’: 0.0, ‘沙’: 0.0, ‘公’: 0.0, ‘转’: 0.0}</p>
<h4>平均召回率</h4>
<p>m_recall : 0.14366647533124086
剔除新多出的关键字 m_recall : 0.3801716303952058</p>
<h5>准确率</h5>
<p>acc_dicts : {‘超’: 0.0, ‘外’: 0.0, ‘应’: 0.0, ‘线’: 0.0, ‘型’: 1.0, ‘19’: 0.0, ‘交’: 0.6666666666666666, ‘G1’: 0.0, ‘7’: 0.0, ‘中’: 0.0, ‘门’: 0.0, ‘行’: 0.7871287128712872, ‘开’: 1.0, ‘停’: 0.0, ‘1Km’: 0.0, ‘兴’: 0.0, ‘小’: 0.8881987577639752, ‘沈’: 0.0, ‘邢’: 0.0, ‘沧’: 0.0, ‘高’: 0.5333333333333333, ‘员’: 0.0, ‘乘’: 0.0, ‘东’: 0.0, ‘向’: 0.0, ‘街’: 0.0, ‘海’: 0.0, ‘丽’: 0.0, ‘夏’: 0.0, ‘延’: 0.0, ‘带’: 0.0, ‘G7’: 0.0, ‘京’: 1.0, ‘二’: 0.0, ‘水’: 0.0, ‘靠’: 0.0, ‘架’: 1.0, ‘慢’: 0.0, ‘10’: 0.0, ‘藏’: 0.0, ‘速’: 0.8, ‘聊’: 0.0, ‘目’: 1.0, ‘快’: 0.0, ‘16’: 0.0, ‘90’: 1.0, ‘混’: 0.0, ‘道’: 0.7916666666666666, ‘机’: 0.0, ‘Km’: 0.0, ‘清’: 0.0, ‘S5’: 0.0, ‘内’: 1.0, ‘方’: 0.0, ‘40’: 0.0, ‘大’: 0.875, ‘60’: 0.8726114649681529, ‘路’: 0.8571428571428571, ‘出’: 0.0, ‘西’: 1.0, ‘桥’: 1.0, ‘新’: 0.0, ‘场’: 0.0, ‘闵’: 0.0, ‘120’: 1.0, ‘用’: 0.0, ‘天’: 1.0, ‘承’: 0.0, ‘城’: 0.0, ‘80’: 0.8866498740554156, ‘货’: 0.0, ‘山’: 0.0, ‘上’: 0.0, ‘厦’: 0.0, ‘辆’: 0.0, ‘50’: 1.0, ‘馆’: 0.0, ‘专’: 0.0, ‘口’: 0.0, ‘站’: 0.0, ‘青’: 0.0, ‘良’: 0.0, ‘区’: 0.0, ‘直’: 0.0, ‘嘉’: 0.0, ‘70’: 0.7894736842105263, ‘G4’: 0.0, ‘车’: 0.8235858101629914, ‘津’: 0.4, ‘一’: 0.3333333333333333, ‘环’: 0.0, ‘客’: 0.6871165644171779, ‘北’: 0.0, ‘右’: 0.0, ‘潘’: 0.0, ‘淮’: 0.0, ‘陶’: 0.0, ‘昆’: 0.0, ‘银’: 0.0, ‘急’: 0.0, ‘2Km’: 1.0, ‘安’: 1.0, ‘100’: 0.8033826638477801, ‘漳’: 0.0, ‘州’: 0.0, ‘都’: 0.0, ‘南’: 0.2777777777777778, ‘浦’: 1.0, ‘临’: 0.0, ‘多’: 0.0, ‘沙’: 0.0, ‘公’: 0.0, ‘转’: 0.0}</p>
<h4>平均准确率</h4>
<p>m_acc : 0.23958467404617645
剔除新多出的关键字 m_acc : 0.502</p>
<h4>优化后文字识别模型</h4>
<h4>召回率</h4>
<p>recall_dicts : {‘超’: 0.5, ‘外’: 0.75, ‘应’: 0.0, ‘线’: 0.0, ‘型’: 0.9, ‘19’: 1.0, ‘交’: 1.0, ‘G1’: 0.6666666666666666, ‘7’: 1.0, ‘中’: 0.18181818181818182, ‘门’: 0.25, ‘行’: 0.9921996879875195, ‘开’: 0.6, ‘停’: 0.6666666666666666, ‘1Km’: 0.0, ‘兴’: 1.0, ‘小’: 0.9765100671140939, ‘沈’: 0.0, ‘邢’: 0.0, ‘沧’: 0.0, ‘高’: 0.8888888888888888, ‘员’: 1.0, ‘乘’: 1.0, ‘东’: 0.0, ‘向’: 0.8888888888888888, ‘街’: 0.6666666666666666, ‘海’: 0.55, ‘丽’: 0.0, ‘夏’: 0.0, ‘延’: 0.0, ‘带’: 0.0, ‘G7’: 0.0, ‘京’: 0.7692307692307693, ‘二’: 0.0, ‘水’: 0.0, ‘靠’: 0.0, ‘架’: 0.8333333333333334, ‘慢’: 0.0, ‘10’: 1.0, ‘藏’: 1.0, ‘速’: 0.5185185185185185, ‘聊’: 0.0, ‘目’: 0.0, ‘快’: 0.5, ‘16’: 1.0, ‘90’: 0.9534883720930233, ‘混’: 0.0, ‘道’: 0.9939759036144579, ‘机’: 1.0, ‘Km’: 0.0, ‘清’: 1.0, ‘S5’: 1.0, ‘内’: 0.25, ‘方’: 1.0, ‘40’: 0.0, ‘大’: 0.9433962264150944, ‘60’: 0.9326424870466321, ‘路’: 0.8181818181818182, ‘出’: 0.6666666666666666, ‘西’: 1.0, ‘桥’: 0.4375, ‘新’: 0.5, ‘场’: 1.0, ‘闵’: 0.7142857142857143, ‘120’: 0.30434782608695654, ‘用’: 1.0, ‘天’: 0.0, ‘承’: 0.0, ‘城’: 0.0, ‘80’: 0.9664948453608248, ‘货’: 0.0, ‘山’: 0.5, ‘上’: 0.9166666666666666, ‘厦’: 0.3333333333333333, ‘辆’: 0.5, ‘50’: 0.5, ‘馆’: 0.0, ‘专’: 1.0, ‘口’: 0.6153846153846154, ‘站’: 0.0, ‘青’: 0.0, ‘良’: 0.3333333333333333, ‘区’: 0.0, ‘直’: 0.0, ‘嘉’: 1.0, ‘70’: 0.9142857142857143, ‘G4’: 0.0, ‘车’: 0.9954022988505747, ‘津’: 1.0, ‘一’: 0.25, ‘环’: 0.6666666666666666, ‘客’: 0.9868421052631579, ‘北’: 0.8823529411764706, ‘右’: 0.6666666666666666, ‘潘’: 0.0, ‘淮’: 0.0, ‘陶’: 0.0, ‘昆’: 0.0, ‘银’: 1.0, ‘急’: 0.0, ‘2Km’: 0.14285714285714285, ‘安’: 0.0, ‘100’: 0.9273182957393483, ‘漳’: 0.0, ‘州’: 0.0, ‘都’: 1.0, ‘南’: 0.75, ‘浦’: 0.42857142857142855, ‘临’: 0.0, ‘多’: 1.0, ‘沙’: 0.0, ‘公’: 0.96875, ‘转’: 1.0}</p>
<h4>平均召回率</h4>
<p>m_recall : 0.49432565844536724
剔除新多出的关键字 m_recall : 0.5636200905459005</p>
<h4>准确率</h4>
<p>acc_dicts : {‘超’: 1.0, ‘外’: 0.75, ‘应’: 0.0, ‘线’: 0.0, ‘型’: 0.75, ‘19’: 1.0, ‘交’: 0.9259259259259259, ‘G1’: 1.0, ‘7’: 0.4166666666666667, ‘中’: 1.0, ‘门’: 1.0, ‘行’: 0.9739663093415007, ‘开’: 0.75, ‘停’: 1.0, ‘1Km’: 0.0, ‘兴’: 0.92, ‘小’: 0.9297124600638977, ‘沈’: 0.0, ‘邢’: 0.0, ‘沧’: 0.0, ‘高’: 0.8888888888888888, ‘员’: 0.8095238095238095, ‘乘’: 0.84, ‘东’: 0.0, ‘向’: 0.8, ‘街’: 1.0, ‘海’: 0.7857142857142857, ‘丽’: 0.0, ‘夏’: 0.0, ‘延’: 0.0, ‘带’: 0.0, ‘G7’: 0.0, ‘京’: 1.0, ‘二’: 0.0, ‘水’: 0.0, ‘靠’: 0.0, ‘架’: 1.0, ‘慢’: 0.0, ‘10’: 0.20833333333333334, ‘藏’: 1.0, ‘速’: 0.875, ‘聊’: 0.0, ‘目’: 0.0, ‘快’: 1.0, ‘16’: 1.0, ‘90’: 0.9534883720930233, ‘混’: 0.0, ‘道’: 0.9428571428571428, ‘机’: 1.0, ‘Km’: 0.0, ‘清’: 1.0, ‘S5’: 1.0, ‘内’: 0.5, ‘方’: 0.9, ‘40’: 0.0, ‘大’: 0.847457627118644, ‘60’: 0.9574468085106383, ‘路’: 0.8181818181818182, ‘出’: 0.8, ‘西’: 1.0, ‘桥’: 0.7, ‘新’: 1.0, ‘场’: 1.0, ‘闵’: 0.7142857142857143, ‘120’: 0.875, ‘用’: 1.0, ‘天’: 0.0, ‘承’: 0.0, ‘城’: 0.0, ‘80’: 0.984251968503937, ‘货’: 0.0, ‘山’: 1.0, ‘上’: 1.0, ‘厦’: 1.0, ‘辆’: 1.0, ‘50’: 1.0, ‘馆’: 0.0, ‘专’: 0.9411764705882353, ‘口’: 0.8888888888888888, ‘站’: 0.0, ‘青’: 0.0, ‘良’: 0.5, ‘区’: 0.0, ‘直’: 0.0, ‘嘉’: 0.5, ‘70’: 0.9795918367346939, ‘G4’: 0.0, ‘车’: 0.9569060773480663, ‘津’: 0.5384615384615384, ‘一’: 1.0, ‘环’: 0.8, ‘客’: 0.9698275862068966, ‘北’: 0.9375, ‘右’: 1.0, ‘潘’: 0.0, ‘淮’: 0.0, ‘陶’: 0.0, ‘昆’: 0.0, ‘银’: 1.0, ‘急’: 0.0, ‘2Km’: 1.0, ‘安’: 0.0, ‘100’: 0.9511568123393316, ‘漳’: 0.0, ‘州’: 0.0, ‘都’: 1.0, ‘南’: 0.46153846153846156, ‘浦’: 0.75, ‘临’: 0.0, ‘多’: 0.9285714285714286, ‘沙’: 0.0, ‘公’: 0.96875, ‘转’: 1.0}</p>
<h4>平均准确率</h4>
<p>m_acc : 0.5636200905459005
剔除新多出的关键字 m_acc : 0.6080598968702589 </p>