Kevin Wang

The network architecture still confused.

1

 @YoungminBaek Hi, the issue #24 u update the network architecture, but it's still confusing. https://github.com/clovaai/CRAFT-pytorch/blob/e332dd8b718e291f51b66ff8f9ef2c98ee4474c8/craft.py#L37-L41 https://github.com/clovaai/CRAFT-pytorch/blob/e332dd8b718e291f51b66ff8f9ef2c98ee4474c8/craft.py#L58-L64 the upconv1 input channel should be 1024+512, so it seems the last stage...

问题描述： 所有DB 的config文件训练部分标签处理 MakeBorderMap和MakeShrinkMap需要调换顺序 https://github.com/PaddlePaddle/PaddleOCR/blob/bde8cad09668939a94fde066ffb556a703d7f7ca/configs/det/det_r50_vd_db.yml#L82-L88 原因分析： 1. MakeShrinkMap会有两个环节会判断标注是否需要ignore_tags： https://github.com/PaddlePaddle/PaddleOCR/blob/bde8cad09668939a94fde066ffb556a703d7f7ca/ppocr/data/imaug/make_shrink_map.py#L56-L59 https://github.com/PaddlePaddle/PaddleOCR/blob/bde8cad09668939a94fde066ffb556a703d7f7ca/ppocr/data/imaug/make_shrink_map.py#L81-L85 2.MakeBorderMap仅会根据ignore_tags进行绘制T图label https://github.com/PaddlePaddle/PaddleOCR/blob/bde8cad09668939a94fde066ffb556a703d7f7ca/ppocr/data/imaug/make_border_map.py#L57-L61 所以先MakeBorderMap再MakeShrinkMap，会存在T图与P图不匹配的情况 3.原作者的repo里是先计算P图label再计算T图label https://github.com/MhLiao/DB/blob/e5a12f5c2f0c2b4a345b5b8392307ef73481d5f6/experiments/seg_detector/base_ic15.yaml#L24-L25

Maybe there is a bug about calculating iou?

1

```Python def iou(box, clusters): """ Calculates the Intersection over Union (IoU) between a box and k clusters. :param box: tuple or array, shifted to the origin (i. e. width and...

utils.py模块中的adjust_lines函数存在使得line_to_line函数报分母为0的BUG

1

https://github.com/chineseocr/table-detect/blob/92488f30ffaf486d29791aab63802beeb1eaca32/utils.py#L180 **稳定触发报错的例子：** 1.ColsLines[i] = (189，5，189，92)；ColsLines[j] = (202, 92, 202, 396)，即(x2,y2) = (189, 92)，(x3,y3) = (202, 92)，r = sqrt((x2, y2), (x3, y3)) < alph 成立 2.会执行newColsLines.append([x2, y2, x3, y3])，但此处显然[189,92,202,92]是一条横线 3.newColsLines列表里混入横线后，运行到line_to_line函数时，只要与[189,92,202,92]计算的另一条横线y值相同，就会导致fit_line时直线系数A为0 https://github.com/chineseocr/table-detect/blob/92488f30ffaf486d29791aab63802beeb1eaca32/utils.py#L228...

并发场景下识别batch推理环节resize_norm_img会使同一张图在不同batch里的处理结果不同导致识别结果会变化

4

https://github.com/PaddlePaddle/PaddleOCR/blob/a67a6fa382cbf55aff12154d0fd3635c3a980b07/tools/infer/predict_rec.py#L110 https://github.com/PaddlePaddle/PaddleOCR/blob/a67a6fa382cbf55aff12154d0fd3635c3a980b07/tools/infer/predict_rec.py#L117 https://github.com/PaddlePaddle/PaddleOCR/blob/a67a6fa382cbf55aff12154d0fd3635c3a980b07/tools/infer/predict_rec.py#L120 假设一张图片A的size是（3， 39， 305） 情况一：batch内的max_wh_ratio即为A的ratio 那么imgW = int(32*(305/39) = 250，即图片A在此batch内的resized_w = 250 情况二：batch内的A的ratio不是最大，假设max_wh_ratio=10 那么imgW = int(32*10) = 320, 此时图片A会执行resized_w = int(math.ceil(imgH * ratio)) ，图片在此batch内的resize_w = 251 在并发的场景下，同一张图片在不同的batch里可能会得到不同的处理结果，从而导致同一张图多次调用会出现不同的识别结果 解决方法：...

… in order to draw Chinese characters properly Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more...

Kevinnunu/add LMDBDumper and TextRecogLMDBConfigGenerator, update prepare_dataset.py script

1

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand...

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand...

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand...

Thanks for your contribution and we appreciate it a lot. The following instructions would make your pull request more healthy and more easily get feedback. If you do not understand...