add PP-ChatOCRv4 and support PDF #2734
Conversation
|
Thanks for your contribution! |
| use_table_recognition=True, | ||
| ) | ||
|
|
||
| # ####[TODO] 增加类别信息 |
| Returns: | ||
| OCRResult: The predicted OCR result with updated dt_boxes. | ||
| """ | ||
| overall_ocr_res = next(self.general_ocr_pipeline(image_array)) |
There was a problem hiding this comment.
这里及之前用next的原因是因为batch_size为1么？
|
|
||
| class TableRecognitionResult(CVResult, HtmlMixin, XlsxMixin): | ||
| class TableRecognitionResult(BaseCVResult, HtmlMixin, XlsxMixin): |
There was a problem hiding this comment.
是否应该将HtmlMixin和XlsxMixin合并为TableMixin，TableMixin提供html、xlsx相关方法。
| class PP_ChatOCR_Pipeline(BasePipeline): | ||
| """PP-ChatOCR Pipeline""" | ||
|
|
||
| entities = ["PP-ChatOCRv3-doc", "PP-ChatOCRv4-doc"] |
There was a problem hiding this comment.
后续ChatOCR迭代的话，仍然会沿用这一逻辑吗，如果后续版本有大的变更，那是否应该分开写？
| layout_parsing_config = config["SubPipelines"]["LayoutParser"] | ||
| self.layout_parsing_pipeline = self.create_pipeline(layout_parsing_config) | ||
| layout_parsing_config = config["SubPipelines"]["LayoutParser"] | ||
| self.layout_parsing_pipeline = self.create_pipeline(layout_parsing_config) | ||
|
|
||
| from .. import create_chat_bot |
| all_table_text_list, | ||
| all_table_html_list, | ||
| all_table_nei_text_list, | ||
| ) = all_visual_info | ||
|
|
||
| final_results = {} | ||
| failed_results = ["大模型调用失败", "未知", "未找到关键信息", "None", ""] |
There was a problem hiding this comment.
看上面的代码，failed_results应该是dict？
There was a problem hiding this comment.
def generate_and_merge_chat_results(self, prompt: str, key_list: list, final_results: dict, failed_results: dict)中，failed_results好像是list
There was a problem hiding this comment.
# print(prompt, llm_result)这里可以删掉
https://github.com/PaddlePaddle/PaddleX/pull/2734/files#diff-6a1b32b7567e8a063375343e919c35d78242c535f36a551f01c4e9f67df77f1aR462
| assert ( | ||
| BaseCVResult.INPUT_IMG_KEY in data | ||
| ), f"`{BaseCVResult.INPUT_IMG_KEY}` is needed, but not found in `{list(data.keys())}`!" | ||
| self._input_img = data.pop("input_img", None) |
There was a problem hiding this comment.
这里建议先不删除吧，我后续提PR删除，因为要配合CV其他模块一起改，其他模块都是依赖self._input_img的，会导致CI挂。
No description provided.