印章检测训练问题

[phb-shiyige-fw]

请提供下述完整信息以便快速定位问题/Please provide the following information to quickly locate the problem

• 系统环境/System Environment：Windows10
• 版本号/Version：Paddle：2.3.2 PaddleOCR：2.6
• 问题相关组件/Related components：det_r50_db++_icdar15.yml
• 运行指令/Command Code：使用db++网络进行两阶段印章检测，检测训练指令
• 完整报错/Complete Error Message：

我按照md中的要求，对500张训练集+200张测试集的标注好的印章进行检测训练，训练epoch为100，训练过程中hmean值最高出现在epoch2中，大约为0.56；此后各个epoch的percision，recall，hmean值都不太理想，请问这是正常的吗？

部分训练日志如下：

[2022/10/24 13:35:54] ppocr INFO: epoch: [29/100], global_step: 4650, lr: 0.005223, loss: 0.301671, loss_shrink_maps: 0.178135, loss_threshold_maps: 0.090503, loss_binary_maps: 0.033389, avg_reader_cost: 1.20758 s, avg_batch_cost: 1.37243 s, avg_samples: 3.0, ips: 2.18591 samples/s, eta: 3:58:40 [2022/10/24 13:36:10] ppocr INFO: epoch: [29/100], global_step: 4660, lr: 0.005219, loss: 0.313256, loss_shrink_maps: 0.187396, loss_threshold_maps: 0.090543, loss_binary_maps: 0.034491, avg_reader_cost: 1.00857 s, avg_batch_cost: 1.22422 s, avg_samples: 3.0, ips: 2.45053 samples/s, eta: 3:58:29 [2022/10/24 13:36:27] ppocr INFO: epoch: [29/100], global_step: 4670, lr: 0.005215, loss: 0.324069, loss_shrink_maps: 0.191917, loss_threshold_maps: 0.099074, loss_binary_maps: 0.035311, avg_reader_cost: 1.04802 s, avg_batch_cost: 1.26288 s, avg_samples: 3.0, ips: 2.37552 samples/s, eta: 3:58:19 [2022/10/24 13:36:43] ppocr INFO: epoch: [29/100], global_step: 4680, lr: 0.005211, loss: 0.305800, loss_shrink_maps: 0.178089, loss_threshold_maps: 0.092302, loss_binary_maps: 0.033797, avg_reader_cost: 0.98907 s, avg_batch_cost: 1.20579 s, avg_samples: 3.0, ips: 2.48799 samples/s, eta: 3:58:08 [2022/10/24 13:36:59] ppocr INFO: epoch: [29/100], global_step: 4690, lr: 0.005207, loss: 0.303913, loss_shrink_maps: 0.179181, loss_threshold_maps: 0.091471, loss_binary_maps: 0.033212, avg_reader_cost: 0.99393 s, avg_batch_cost: 1.20915 s, avg_samples: 3.0, ips: 2.48108 samples/s, eta: 3:57:56 [2022/10/24 13:37:15] ppocr INFO: epoch: [29/100], global_step: 4700, lr: 0.005203, loss: 0.327307, loss_shrink_maps: 0.191015, loss_threshold_maps: 0.102151, loss_binary_maps: 0.035830, avg_reader_cost: 1.00747 s, avg_batch_cost: 1.22321 s, avg_samples: 3.0, ips: 2.45256 samples/s, eta: 3:57:45 [2022/10/24 13:37:31] ppocr INFO: epoch: [29/100], global_step: 4710, lr: 0.005199, loss: 0.317068, loss_shrink_maps: 0.184073, loss_threshold_maps: 0.097330, loss_binary_maps: 0.034826, avg_reader_cost: 1.01170 s, avg_batch_cost: 1.22742 s, avg_samples: 3.0, ips: 2.44415 samples/s, eta: 3:57:34 [2022/10/24 13:37:48] ppocr INFO: epoch: [29/100], global_step: 4720, lr: 0.005195, loss: 0.308528, loss_shrink_maps: 0.183476, loss_threshold_maps: 0.093512, loss_binary_maps: 0.032611, avg_reader_cost: 1.02460 s, avg_batch_cost: 1.24009 s, avg_samples: 3.0, ips: 2.41919 samples/s, eta: 3:57:23 [2022/10/24 13:38:04] ppocr INFO: epoch: [29/100], global_step: 4730, lr: 0.005191, loss: 0.309441, loss_shrink_maps: 0.180652, loss_threshold_maps: 0.095098, loss_binary_maps: 0.033086, avg_reader_cost: 1.00472 s, avg_batch_cost: 1.21924 s, avg_samples: 3.0, ips: 2.46056 samples/s, eta: 3:57:12 [2022/10/24 13:38:19] ppocr INFO: epoch: [29/100], global_step: 4740, lr: 0.005187, loss: 0.308559, loss_shrink_maps: 0.178189, loss_threshold_maps: 0.095626, loss_binary_maps: 0.034017, avg_reader_cost: 0.97421 s, avg_batch_cost: 1.18963 s, avg_samples: 3.0, ips: 2.52179 samples/s, eta: 3:57:00 [2022/10/24 13:38:36] ppocr INFO: epoch: [29/100], global_step: 4750, lr: 0.005184, loss: 0.327761, loss_shrink_maps: 0.193155, loss_threshold_maps: 0.097956, loss_binary_maps: 0.035140, avg_reader_cost: 1.00805 s, avg_batch_cost: 1.22316 s, avg_samples: 3.0, ips: 2.45266 samples/s, eta: 3:56:49 [2022/10/24 13:38:52] ppocr INFO: epoch: [29/100], global_step: 4760, lr: 0.005180, loss: 0.348365, loss_shrink_maps: 0.206605, loss_threshold_maps: 0.103290, loss_binary_maps: 0.037214, avg_reader_cost: 1.08738 s, avg_batch_cost: 1.30369 s, avg_samples: 3.0, ips: 2.30116 samples/s, eta: 3:56:40 [2022/10/24 13:39:10] ppocr INFO: epoch: [29/100], global_step: 4770, lr: 0.005176, loss: 0.350313, loss_shrink_maps: 0.207048, loss_threshold_maps: 0.101927, loss_binary_maps: 0.037214, avg_reader_cost: 1.10052 s, avg_batch_cost: 1.31639 s, avg_samples: 3.0, ips: 2.27896 samples/s, eta: 3:56:31 [2022/10/24 13:39:25] ppocr INFO: epoch: [29/100], global_step: 4780, lr: 0.005172, loss: 0.304381, loss_shrink_maps: 0.177115, loss_threshold_maps: 0.094480, loss_binary_maps: 0.032863, avg_reader_cost: 0.97998 s, avg_batch_cost: 1.19583 s, avg_samples: 3.0, ips: 2.50873 samples/s, eta: 3:56:19 [2022/10/24 13:39:41] ppocr INFO: epoch: [29/100], global_step: 4790, lr: 0.005168, loss: 0.303450, loss_shrink_maps: 0.176967, loss_threshold_maps: 0.094591, loss_binary_maps: 0.032507, avg_reader_cost: 0.99098 s, avg_batch_cost: 1.20764 s, avg_samples: 3.0, ips: 2.48419 samples/s, eta: 3:56:08 [2022/10/24 13:39:57] ppocr INFO: epoch: [29/100], global_step: 4800, lr: 0.005164, loss: 0.302331, loss_shrink_maps: 0.177497, loss_threshold_maps: 0.095178, loss_binary_maps: 0.032363, avg_reader_cost: 0.99304 s, avg_batch_cost: 1.20874 s, avg_samples: 3.0, ips: 2.48192 samples/s, eta: 3:55:56 [2022/10/24 13:41:01] ppocr INFO: cur metric, precision: 0.15878070973612374, recall: 0.5065312046444121, hmean: 0.2417734672670592, fps: 3.576573946067822 [2022/10/24 13:41:01] ppocr INFO: best metric, hmean: 0.5654610290371879, is_float16: False, precision: 0.43563579277864994, recall: 0.8055152394775036, fps: 3.5958188240972992, best_epoch: 2 [2022/10/24 13:41:11] ppocr INFO: epoch: [29/100], global_step: 4810, lr: 0.005160, loss: 0.316871, loss_shrink_maps: 0.182529, loss_threshold_maps: 0.095351, loss_binary_maps: 0.034103, avg_reader_cost: 0.41929 s, avg_batch_cost: 0.63451 s, avg_samples: 3.0, ips: 4.72806 samples/s, eta: 3:55:31 [2022/10/24 13:41:19] ppocr INFO: save model in ./output/det_r50_icdar15/latest [2022/10/24 13:41:29] ppocr INFO: epoch: [30/100], global_step: 4820, lr: 0.005156, loss: 0.321667, loss_shrink_maps: 0.185223, loss_threshold_maps: 0.096997, loss_binary_maps: 0.034589, avg_reader_cost: 1.23911 s, avg_batch_cost: 1.40793 s, avg_samples: 3.0, ips: 2.13078 samples/s, eta: 3:55:23 [2022/10/24 13:41:46] ppocr INFO: epoch: [30/100], global_step: 4830, lr: 0.005152, loss: 0.314509, loss_shrink_maps: 0.181742, loss_threshold_maps: 0.096374, loss_binary_maps: 0.034034, avg_reader_cost: 1.00902 s, avg_batch_cost: 1.22453 s, avg_samples: 3.0, ips: 2.44991 samples/s, eta: 3:55:12 [2022/10/24 13:42:02] ppocr INFO: epoch: [30/100], global_step: 4840, lr: 0.005148, loss: 0.303421, loss_shrink_maps: 0.182769, loss_threshold_maps: 0.092221, loss_binary_maps: 0.033133, avg_reader_cost: 1.04194 s, avg_batch_cost: 1.25689 s, avg_samples: 3.0, ips: 2.38685 samples/s, eta: 3:55:01 [2022/10/24 13:42:18] ppocr INFO: epoch: [30/100], global_step: 4850, lr: 0.005145, loss: 0.309537, loss_shrink_maps: 0.184302, loss_threshold_maps: 0.096921, loss_binary_maps: 0.034097, avg_reader_cost: 1.01710 s, avg_batch_cost: 1.23243 s, avg_samples: 3.0, ips: 2.43422 samples/s, eta: 3:54:50 [2022/10/24 13:42:35] ppocr INFO: epoch: [30/100], global_step: 4860, lr: 0.005141, loss: 0.310129, loss_shrink_maps: 0.182205, loss_threshold_maps: 0.097491, loss_binary_maps: 0.034097, avg_reader_cost: 1.08980 s, avg_batch_cost: 1.30441 s, avg_samples: 3.0, ips: 2.29988 samples/s, eta: 3:54:41 [2022/10/24 13:42:51] ppocr INFO: epoch: [30/100], global_step: 4870, lr: 0.005137, loss: 0.317974, loss_shrink_maps: 0.190266, loss_threshold_maps: 0.093979, loss_binary_maps: 0.034426, avg_reader_cost: 1.02116 s, avg_batch_cost: 1.23544 s, avg_samples: 3.0, ips: 2.42828 samples/s, eta: 3:54:30 [2022/10/24 13:43:08] ppocr INFO: epoch: [30/100], global_step: 4880, lr: 0.005133, loss: 0.301216, loss_shrink_maps: 0.175989, loss_threshold_maps: 0.092204, loss_binary_maps: 0.033623, avg_reader_cost: 1.00761 s, avg_batch_cost: 1.22200 s, avg_samples: 3.0, ips: 2.45499 samples/s, eta: 3:54:19 [2022/10/24 13:43:24] ppocr INFO: epoch: [30/100], global_step: 4890, lr: 0.005129, loss: 0.306748, loss_shrink_maps: 0.180882, loss_threshold_maps: 0.091365, loss_binary_maps: 0.033596, avg_reader_cost: 1.03110 s, avg_batch_cost: 1.24503 s, avg_samples: 3.0, ips: 2.40958 samples/s, eta: 3:54:08 [2022/10/24 13:43:40] ppocr INFO: epoch: [30/100], global_step: 4900, lr: 0.005125, loss: 0.306177, loss_shrink_maps: 0.182581, loss_threshold_maps: 0.090128, loss_binary_maps: 0.033506, avg_reader_cost: 0.96897 s, avg_batch_cost: 1.18308 s, avg_samples: 3.0, ips: 2.53575 samples/s, eta: 3:53:56 [2022/10/24 13:43:56] ppocr INFO: epoch: [30/100], global_step: 4910, lr: 0.005121, loss: 0.302035, loss_shrink_maps: 0.178804, loss_threshold_maps: 0.090128, loss_binary_maps: 0.032373, avg_reader_cost: 0.99985 s, avg_batch_cost: 1.21400 s, avg_samples: 3.0, ips: 2.47116 samples/s, eta: 3:53:45 [2022/10/24 13:44:12] ppocr INFO: epoch: [30/100], global_step: 4920, lr: 0.005117, loss: 0.311466, loss_shrink_maps: 0.184811, loss_threshold_maps: 0.093279, loss_binary_maps: 0.032925, avg_reader_cost: 1.07587 s, avg_batch_cost: 1.29036 s, avg_samples: 3.0, ips: 2.32494 samples/s, eta: 3:53:35 [2022/10/24 13:44:29] ppocr INFO: epoch: [30/100], global_step: 4930, lr: 0.005113, loss: 0.325775, loss_shrink_maps: 0.194057, loss_threshold_maps: 0.096067, loss_binary_maps: 0.034653, avg_reader_cost: 1.01361 s, avg_batch_cost: 1.22779 s, avg_samples: 3.0, ips: 2.44341 samples/s, eta: 3:53:24 [2022/10/24 13:44:45] ppocr INFO: epoch: [30/100], global_step: 4940, lr: 0.005109, loss: 0.329996, loss_shrink_maps: 0.199998, loss_threshold_maps: 0.098940, loss_binary_maps: 0.035501, avg_reader_cost: 1.05791 s, avg_batch_cost: 1.27183 s, avg_samples: 3.0, ips: 2.35880 samples/s, eta: 3:53:14 [2022/10/24 13:45:02] ppocr INFO: epoch: [30/100], global_step: 4950, lr: 0.005105, loss: 0.346556, loss_shrink_maps: 0.205951, loss_threshold_maps: 0.102338, loss_binary_maps: 0.036919, avg_reader_cost: 1.03770 s, avg_batch_cost: 1.25221 s, avg_samples: 3.0, ips: 2.39575 samples/s, eta: 3:53:04 [2022/10/24 13:45:18] ppocr INFO: epoch: [30/100], global_step: 4960, lr: 0.005102, loss: 0.327894, loss_shrink_maps: 0.198065, loss_threshold_maps: 0.099057, loss_binary_maps: 0.037082, avg_reader_cost: 1.01615 s, avg_batch_cost: 1.23055 s, avg_samples: 3.0, ips: 2.43793 samples/s, eta: 3:52:53 [2022/10/24 13:45:34] ppocr INFO: epoch: [30/100], global_step: 4970, lr: 0.005098, loss: 0.324318, loss_shrink_maps: 0.190305, loss_threshold_maps: 0.094497, loss_binary_maps: 0.035356, avg_reader_cost: 1.01267 s, avg_batch_cost: 1.22705 s, avg_samples: 3.0, ips: 2.44489 samples/s, eta: 3:52:41 [2022/10/24 13:45:50] ppocr INFO: epoch: [30/100], global_step: 4980, lr: 0.005094, loss: 0.297931, loss_shrink_maps: 0.176329, loss_threshold_maps: 0.088850, loss_binary_maps: 0.034015, avg_reader_cost: 1.01007 s, avg_batch_cost: 1.22420 s, avg_samples: 3.0, ips: 2.45058 samples/s, eta: 3:52:30 [2022/10/24 13:45:51] ppocr INFO: save model in ./output/det_r50_icdar15/latest [2022/10/24 13:45:51] ppocr INFO: save model in ./output/det_r50_icdar15/iter_epoch_30 [2022/10/24 13:46:07] ppocr INFO: epoch: [31/100], global_step: 4990, lr: 0.005090, loss: 0.306258, loss_shrink_maps: 0.179481, loss_threshold_maps: 0.089624, loss_binary_maps: 0.034245, avg_reader_cost: 1.15936 s, avg_batch_cost: 1.31723 s, avg_samples: 3.0, ips: 2.27750 samples/s, eta: 3:52:20 [2022/10/24 13:46:23] ppocr INFO: epoch: [31/100], global_step: 5000, lr: 0.005086, loss: 0.303210, loss_shrink_maps: 0.178812, loss_threshold_maps: 0.090426, loss_binary_maps: 0.032483, avg_reader_cost: 1.01759 s, avg_batch_cost: 1.23237 s, avg_samples: 3.0, ips: 2.43434 samples/s, eta: 3:52:09 [2022/10/24 13:47:27] ppocr INFO: cur metric, precision: 0.20061887570912842, recall: 0.5645863570391872, hmean: 0.2960426179604262, fps: 3.582826279606882 [2022/10/24 13:47:27] ppocr INFO: best metric, hmean: 0.5654610290371879, is_float16: False, precision: 0.43563579277864994, recall: 0.8055152394775036, fps: 3.5958188240972992, best_epoch: 2 [2022/10/24 13:47:37] ppocr INFO: epoch: [31/100], global_step: 5010, lr: 0.005082, loss: 0.301825, loss_shrink_maps: 0.176758, loss_threshold_maps: 0.090151, loss_binary_maps: 0.032856, avg_reader_cost: 0.39577 s, avg_batch_cost: 0.61064 s, avg_samples: 3.0, ips: 4.91289 samples/s, eta: 3:51:43 [2022/10/24 13:47:52] ppocr INFO: epoch: [31/100], global_step: 5020, lr: 0.005078, loss: 0.305158, loss_shrink_maps: 0.179033, loss_threshold_maps: 0.094230, loss_binary_maps: 0.033468, avg_reader_cost: 0.94724 s, avg_batch_cost: 1.16251 s, avg_samples: 3.0, ips: 2.58063 samples/s, eta: 3:51:31 [2022/10/24 13:48:09] ppocr INFO: epoch: [31/100], global_step: 5030, lr: 0.005074, loss: 0.312290, loss_shrink_maps: 0.182317, loss_threshold_maps: 0.095003, loss_binary_maps: 0.033824, avg_reader_cost: 1.03846 s, avg_batch_cost: 1.25339 s, avg_samples: 3.0, ips: 2.39351 samples/s, eta: 3:51:20 [2022/10/24 13:48:24] ppocr INFO: epoch: [31/100], global_step: 5040, lr: 0.005070, loss: 0.319863, loss_shrink_maps: 0.188166, loss_threshold_maps: 0.095565, loss_binary_maps: 0.035005, avg_reader_cost: 0.98116 s, avg_batch_cost: 1.19556 s, avg_samples: 3.0, ips: 2.50928 samples/s, eta: 3:51:09 [2022/10/24 13:48:40] ppocr INFO: epoch: [31/100], global_step: 5050, lr: 0.005066, loss: 0.318157, loss_shrink_maps: 0.186115, loss_threshold_maps: 0.093282, loss_binary_maps: 0.034367, avg_reader_cost: 0.90928 s, avg_batch_cost: 1.12387 s, avg_samples: 3.0, ips: 2.66934 samples/s, eta: 3:50:55 [2022/10/24 13:48:55] ppocr INFO: epoch: [31/100], global_step: 5060, lr: 0.005062, loss: 0.320742, loss_shrink_maps: 0.191787, loss_threshold_maps: 0.092695, loss_binary_maps: 0.034874, avg_reader_cost: 0.93616 s, avg_batch_cost: 1.15057 s, avg_samples: 3.0, ips: 2.60739 samples/s, eta: 3:50:42 [2022/10/24 13:49:11] ppocr INFO: epoch: [31/100], global_step: 5070, lr: 0.005059, loss: 0.352797, loss_shrink_maps: 0.216243, loss_threshold_maps: 0.101480, loss_binary_maps: 0.038967, avg_reader_cost: 0.97962 s, avg_batch_cost: 1.19340 s, avg_samples: 3.0, ips: 2.51383 samples/s, eta: 3:50:30 [2022/10/24 13:49:26] ppocr INFO: epoch: [31/100], global_step: 5080, lr: 0.005055, loss: 0.326052, loss_shrink_maps: 0.192028, loss_threshold_maps: 0.094244, loss_binary_maps: 0.035436, avg_reader_cost: 0.93945 s, avg_batch_cost: 1.15377 s, avg_samples: 3.0, ips: 2.60018 samples/s, eta: 3:50:17 [2022/10/24 13:49:42] ppocr INFO: epoch: [31/100], global_step: 5090, lr: 0.005051, loss: 0.316351, loss_shrink_maps: 0.189200, loss_threshold_maps: 0.094608, loss_binary_maps: 0.034543, avg_reader_cost: 0.97763 s, avg_batch_cost: 1.19296 s, avg_samples: 3.0, ips: 2.51475 samples/s, eta: 3:50:06 [2022/10/24 13:49:58] ppocr INFO: epoch: [31/100], global_step: 5100, lr: 0.005047, loss: 0.307062, loss_shrink_maps: 0.183655, loss_threshold_maps: 0.091638, loss_binary_maps: 0.034293, avg_reader_cost: 0.96019 s, avg_batch_cost: 1.17427 s, avg_samples: 3.0, ips: 2.55477 samples/s, eta: 3:49:53 [2022/10/24 13:50:14] ppocr INFO: epoch: [31/100], global_step: 5110, lr: 0.005043, loss: 0.305575, loss_shrink_maps: 0.181562, loss_threshold_maps: 0.091365, loss_binary_maps: 0.034225, avg_reader_cost: 1.02860 s, avg_batch_cost: 1.24359 s, avg_samples: 3.0, ips: 2.41238 samples/s, eta: 3:49:42 [2022/10/24 13:50:30] ppocr INFO: epoch: [31/100], global_step: 5120, lr: 0.005039, loss: 0.315851, loss_shrink_maps: 0.185708, loss_threshold_maps: 0.092877, loss_binary_maps: 0.034747, avg_reader_cost: 0.98041 s, avg_batch_cost: 1.19492 s, avg_samples: 3.0, ips: 2.51062 samples/s, eta: 3:49:31 [2022/10/24 13:50:46] ppocr INFO: epoch: [31/100], global_step: 5130, lr: 0.005035, loss: 0.297900, loss_shrink_maps: 0.174107, loss_threshold_maps: 0.090991, loss_binary_maps: 0.032588, avg_reader_cost: 0.98516 s, avg_batch_cost: 1.19978 s, avg_samples: 3.0, ips: 2.50047 samples/s, eta: 3:49:19 [2022/10/24 13:51:02] ppocr INFO: epoch: [31/100], global_step: 5140, lr: 0.005031, loss: 0.297638, loss_shrink_maps: 0.174107, loss_threshold_maps: 0.091328, loss_binary_maps: 0.032120, avg_reader_cost: 0.98741 s, avg_batch_cost: 1.20260 s, avg_samples: 3.0, ips: 2.49459 samples/s, eta: 3:49:07

还有请问 能够给一份具体的检测参数配置文件吗？就是使用db++检测印章的配置文件，因为教程中只要求改数据集地址。

[LDOUBLEV]

最近repo里代码有一些修改，如果是检测弯曲框，需要在后处理参数中设置为det_box_type: 'poly'

https://github.com/PaddlePaddle/PaddleOCR/blob/6228d1ee1364845cd5f4af92f150b2cb0ebaa539/configs/det/det_r50_db%2B%2B_icdar15.yml#L57

改为：det_box_type: 'poly'

让后处理走这个if分支： https://github.com/PaddlePaddle/PaddleOCR/blob/6228d1ee1364845cd5f4af92f150b2cb0ebaa539/ppocr/postprocess/db_postprocess.py#L236

[phb-shiyige-fw]

最近repo里代码有一些修改，如果是检测弯曲框，需要在后处理参数中设置为det_box_type: 'poly'

https://github.com/PaddlePaddle/PaddleOCR/blob/6228d1ee1364845cd5f4af92f150b2cb0ebaa539/configs/det/det_r50_db%2B%2B_icdar15.yml#L57

改为：det_box_type: 'poly'

让后处理走这个if分支：

https://github.com/PaddlePaddle/PaddleOCR/blob/6228d1ee1364845cd5f4af92f150b2cb0ebaa539/ppocr/postprocess/db_postprocess.py#L236

你好 之前我看FAQ中是使用use_polygon=true 现在是不能使用了吗？

[xialei2821212670]

最近repo里代码有一些修改，如果是检测弯曲框，需要在后处理参数中设置为det_box_type: 'poly' https://github.com/PaddlePaddle/PaddleOCR/blob/6228d1ee1364845cd5f4af92f150b2cb0ebaa539/configs/det/det_r50_db%2B%2B_icdar15.yml#L57

改为：det_box_type: 'poly' 让后处理走这个如果分支： https://github.com/PaddlePaddle/PaddleOCR/blob/6228d1ee1364845cd5f4af92f150b2cb0ebaa539/ppocr/postprocess/db_postprocess.py#L236

你好之前我看FAQ中是使用use_polygon=true现在是不能使用了吗？

可以使用，预测的时候改成poly,就行。部署服务没问题