Add rec_sar #3798
Add rec_sar #3798
Conversation
|
Thanks for your contribution! |
|
|
| algorithm: SAR | ||
| Transform: | ||
| Backbone: | ||
| name: ResNet31 |
There was a problem hiding this comment.
resnet31是作者新改的一个网络结构，和常用的不太一样
| channel_first: False | ||
| - SARLabelEncode: # Class handling label | ||
| - SARRecResizeImg: | ||
| image_shape: [3, 48, 48, 160] # h:48 w:[48,160] |
There was a problem hiding this comment.
后两维是宽度的范围，宽度是变长的
configs/rec/rec_r31_sar.yml
Outdated
| dataset: | ||
| name: SimpleDataSet | ||
| delimiter: ' ' | ||
| label_file_list: ['/paddle/data/concat_data/icdar_2013_train20.txt', '/paddle/data/concat_data/icdar_2015_train20.txt', '/paddle/data/concat_data/coco_text_train20.txt', '/paddle/data/concat_data/IIIt5k_train20.txt', '/paddle/data/concat_data/SynthAdd_train.txt', '/paddle/data/concat_data/SynthText_train.txt', '/paddle/data/concat_data/Syn90k_train.txt'] |
There was a problem hiding this comment.
建议把数据路径替换成 train_data/train_list.txt
在文档里说明训练需要用到哪些数据，有什么不同
There was a problem hiding this comment.
那就是把这几个txt合成一个吗？
configs/rec/rec_r31_sar.yml
Outdated
| Eval: | ||
| dataset: | ||
| name: LMDBDataSet | ||
| data_dir: /paddle/data/ocr_data/evaluation/ |
There was a problem hiding this comment.
不要用绝对路径，指向相对路径，让用户可以很方便跑通，参考其他算法的配置文件。上面train同理
| comp = re.compile('[^A-Z^a-z^0-9^\u4e00-\u9fa5]') | ||
| text = text.lower() | ||
| text = comp.sub('', text) | ||
| result_list.append((text, np.mean(conf_list))) |
There was a problem hiding this comment.
建议写成可配置，或者写在metric计算里。不要影响正常的infer结果。这样如果替换成中文训练，需要手动修改过滤条件，不是很灵活。
There was a problem hiding this comment.
之前不是说不要改metric吗？所以写在了后处理里。那可以加一个参数来控制？
ppocr/utils/dict90.txt
Outdated
| ] | ||
| _ | ||
| ` | ||
| ~ |
There was a problem hiding this comment.
如果用94字符的字典训练，精度会下降吗？ 没区别的话建议修改成 character_type == "EN_symbol"，不要提交重复性过高的字典。
There was a problem hiding this comment.
好的，这个前面是为了和竞品配置保持一样
add recognition model SAR