add tablemaster to tipc

27181149 · 文幕地方 · 52d5b545 · 27181149 · 27181149 · 27181149
10 changed file
--- a/doc/doc_ch/algorithm.md
+++ b/doc/doc_ch/algorithm.md
@@ -5,6 +5,7 @@ PaddleOCR将**持续新增**支持OCR领域前沿算法与模型，已支持的
 - [文本检测算法](./algorithm_overview.md#11-%E6%96%87%E6%9C%AC%E6%A3%80%E6%B5%8B%E7%AE%97%E6%B3%95)
 - [文本识别算法](./algorithm_overview.md#12-%E6%96%87%E6%9C%AC%E8%AF%86%E5%88%AB%E7%AE%97%E6%B3%95)
 - [端到端算法](./algorithm_overview.md#2-%E6%96%87%E6%9C%AC%E8%AF%86%E5%88%AB%E7%AE%97%E6%B3%95)
+- [表格识别]](./algorithm_overview.md#3-%E8%A1%A8%E6%A0%BC%E8%AF%86%E5%88%AB%E7%AE%97%E6%B3%95)
 **欢迎广大开发者合作共建，贡献更多算法，合入有奖🎁！具体可查看[社区常规赛](https://github.com/PaddlePaddle/PaddleOCR/issues/4982)。**

--- a/doc/doc_ch/algorithm_overview.md
+++ b/doc/doc_ch/algorithm_overview.md
 # OCR算法
- [1. 两阶段算法](#1-两阶段算法)
+- [1. 两阶段算法](#1)
-  - [1.1 文本检测算法](#11-文本检测算法)
+  - [1.1 文本检测算法](#11)
-  - [1.2 文本识别算法](#12-文本识别算法)
+  - [1.2 文本识别算法](#12)
- [2. 端到端算法](#2-端到端算法)
+- [2. 端到端算法](#2)
- [3. 表格识别算法](#3-表格识别算法)
+- [3. 表格识别算法](#3)
 本文给出了PaddleOCR已支持的OCR算法列表，以及每个算法在**英文公开数据集**上的模型和指标，主要用于算法简介和算法性能对比，更多包括中文在内的其他数据集上的模型请参考[PP-OCR v2.0 系列模型下载](./models_list.md)。
@@ -98,6 +98,8 @@
 已支持的端到端OCR算法列表（戳链接获取使用教程）：
 - [x]  [PGNet](./algorithm_e2e_pgnet.md)
+<a name="3"></a>
 ## 3. 表格识别算法
 已支持的表格识别算法列表（戳链接获取使用教程）：

--- a/doc/doc_en/algorithm_en.md
+++ b/doc/doc_en/algorithm_en.md
@@ -6,5 +6,6 @@ PaddleOCR will add cutting-edge OCR algorithms and models continuously. Check ou
 - [text detection algorithms](./algorithm_overview_en.md#11)
 - [text recognition algorithms](./algorithm_overview_en.md#12)
 - [end-to-end algorithms](./algorithm_overview_en.md#2)
+- [table recognition algorithms](./algorithm_overview_en.md#3)
 Developers are welcome to contribute more algorithms! Please refer to [add new algorithm](./add_new_algorithm_en.md) guideline.
--- a/doc/doc_en/algorithm_overview_en.md
+++ b/doc/doc_en/algorithm_overview_en.md
 # OCR Algorithms
- [1. Two-stage Algorithms](#1-two-stage-algorithms)
+- [1. Two-stage Algorithms](#1)
-  - [1.1 Text Detection Algorithms](#11-text-detection-algorithms)
+  - [1.1 Text Detection Algorithms](#11)
-  - [1.2 Text Recognition Algorithms](#12-text-recognition-algorithms)
+  - [1.2 Text Recognition Algorithms](#12)
- [2. End-to-end Algorithms](#2-end-to-end-algorithms)
+- [2. End-to-end Algorithms](#2)
- [3. Table Recognition Algorithms](#3-table-recognition-algorithms)
+- [3. Table Recognition Algorithms](#3)
 This tutorial lists the OCR algorithms supported by PaddleOCR, as well as the models and metrics of each algorithm on **English public datasets**. It is mainly used for algorithm introduction and algorithm performance comparison. For more models on other datasets including Chinese, please refer to [PP-OCR v2.0 models list](./models_list_en.md).

--- a/ppstructure/table/predict_structure.py
+++ b/ppstructure/table/predict_structure.py
@@ -118,7 +118,7 @@ class TableStructurer(object):
            '<html>', '<body>', '<table>'
        ] + structure_str_list + ['</table>', '</body>', '</html>']
        elapse = time.time() - starttime
-        return structure_str_list, bbox_list, elapse
+        return (structure_str_list, bbox_list), elapse
 def main(args):
@@ -138,8 +138,8 @@ def main(args):
            if img is None:
                logger.info("error in loading image:{}".format(image_file))
                continue
-            structure_str_list, bbox_list, elapse = table_structurer(img)
+            structure_res, elapse = table_structurer(img)
+            structure_str_list, bbox_list = structure_res
            bbox_list_str = json.dumps(bbox_list.tolist())
            logger.info("result: {}, {}".format(structure_str_list,
                                                bbox_list_str))

--- a/test_tipc/configs/en_table_structure/table_mv3.yml
+++ b/test_tipc/configs/en_table_structure/table_mv3.yml
 Global:
  use_gpu: true
-  epoch_num: 10
+  epoch_num: 400
  log_smooth_window: 20
  print_batch_step: 5
  save_model_dir: ./output/table_mv3/
-  save_epoch_step: 3
+  save_epoch_step: 400
  # evaluation is run every 400 iterations after the 0th iteration
  eval_batch_step: [0, 400]
  cal_metric_during_train: True
@@ -12,13 +12,12 @@ Global:
  checkpoints:
  save_inference_dir:
  use_visualdl: False
-  infer_img: doc/table/table.jpg
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: output/table_mv3
  # for data or label process
  character_dict_path: ppocr/utils/dict/table_structure_dict.txt
  character_type: en
-  max_text_length: 100
+  max_text_length: 800
-  max_elem_length: 800
-  max_cell_num: 500
  infer_mode: False
  process_total_num: 0
  process_cut_num: 0
@@ -44,11 +43,8 @@ Architecture:
  Head:
    name: TableAttentionHead
    hidden_size: 256
-    l2_decay: 0.00001
    loc_type: 2
-    max_text_length: 100
+    max_text_length: 800
-    max_elem_length: 800
-    max_cell_num: 500
 Loss:
  name: TableAttentionLoss
@@ -61,28 +57,34 @@ PostProcess:
 Metric:
  name: TableMetric
  main_indicator: acc
+  compute_bbox_metric: false # cost many time, set False for training
 Train:
  dataset:
    name: PubTabDataSet
    data_dir: ./train_data/pubtabnet/train
-    label_file_path: ./train_data/pubtabnet/train.jsonl
+    label_file_list: [./train_data/pubtabnet/train.jsonl]
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: False
+      - TableBoxEncode:
      - ResizeTableImage:
          max_len: 488
-      - TableLabelEncode:
      - NormalizeImage:
          scale: 1./255.
          mean: [0.485, 0.456, 0.406]
          std: [0.229, 0.224, 0.225]
          order: 'hwc'
      - PaddingTableImage:
+          size: [488, 488]
      - ToCHWImage:
      - KeepKeys:
-          keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
  loader:
    shuffle: True
    batch_size_per_card: 32
@@ -93,23 +95,28 @@ Eval:
  dataset:
    name: PubTabDataSet
    data_dir: ./train_data/pubtabnet/test/
-    label_file_path: ./train_data/pubtabnet/test.jsonl
+    label_file_list: [./train_data/pubtabnet/test.jsonl]
    transforms:
      - DecodeImage: # load image
          img_mode: BGR
          channel_first: False
+      - TableLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: False
+          replace_empty_cell_token: False
+      - TableBoxEncode:
      - ResizeTableImage:
          max_len: 488
-      - TableLabelEncode:
      - NormalizeImage:
          scale: 1./255.
          mean: [0.485, 0.456, 0.406]
          std: [0.229, 0.224, 0.225]
          order: 'hwc'
      - PaddingTableImage:
+          size: [488, 488]
      - ToCHWImage:
      - KeepKeys:
-          keep_keys: ['image', 'structure', 'bbox_list', 'sp_tokens', 'bbox_list_mask']
+          keep_keys: [ 'image', 'structure', 'bboxes', 'bbox_masks', 'shape' ]
  loader:
    shuffle: False
    drop_last: False

--- a/test_tipc/configs/table_master/table_master.yml
+++ b/test_tipc/configs/table_master/table_master.yml
+Global:
+  use_gpu: true
+  epoch_num: 17
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/table_master/
+  save_epoch_step: 17
+  eval_batch_step: [0,  6259]
+  cal_metric_during_train: true
+  pretrained_model: null
+  checkpoints: 
+  save_inference_dir: output/table_master/infer
+  use_visualdl: false
+  infer_img: ppstructure/docs/table/table.jpg
+  save_res_path: ./output/table_master
+  character_dict_path: ppocr/utils/dict/table_master_structure_dict.txt
+  infer_mode: false
+  max_text_length: 500
+  process_total_num: 0
+  process_cut_num: 0
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: MultiStepDecay
+    learning_rate: 0.001
+    milestones: [12, 15]
+    gamma: 0.1
+    warmup_epoch: 0.02
+  regularizer:
+    name: L2
+    factor: 0.0
+Architecture:
+  model_type: table
+  algorithm: TableMaster
+  Backbone:
+    name: TableResNetExtra
+    gcb_config:
+      ratio: 0.0625
+      headers: 1
+      att_scale: False
+      fusion_type: channel_add
+      layers: [False, True, True, True]
+    layers: [1,2,5,3]
+  Head:
+    name: TableMasterHead
+    hidden_size: 512
+    headers: 8
+    dropout: 0
+    d_ff: 2024
+    max_text_length: 500
+Loss:
+  name: TableMasterLoss
+  ignore_index: 42 # set to len of dict + 3
+PostProcess:
+  name: TableMasterLabelDecode
+  box_shape: pad
+Metric:
+  name: TableMetric
+  main_indicator: acc
+  compute_bbox_metric: False
+Train:
+  dataset:
+    name: PubTabDataSet
+    data_dir: ./train_data/pubtabnet/train
+    label_file_list: [./train_data/pubtabnet/train.jsonl]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: True
+          replace_empty_cell_token: True
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          use_xywh: True
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
+  loader:
+    shuffle: True
+    batch_size_per_card: 10
+    drop_last: True
+    num_workers: 8
+Eval:
+  dataset:
+    name: PubTabDataSet
+    data_dir: ./train_data/pubtabnet/test/
+    label_file_list: [./train_data/pubtabnet/test.jsonl]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - TableMasterLabelEncode:
+          learn_empty_box: False
+          merge_no_span_structure: True
+          replace_empty_cell_token: True
+      - ResizeTableImage:
+          max_len: 480
+          resize_bboxes: True
+      - PaddingTableImage:
+          size: [480, 480]
+      - TableBoxEncode:
+          use_xywh: True
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.5, 0.5, 0.5]
+          std: [0.5, 0.5, 0.5]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, structure, bboxes, bbox_masks, shape]
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 10
+    num_workers: 8
\ No newline at end of file
--- a/test_tipc/configs/table_master/train_infer_python.txt
+++ b/test_tipc/configs/table_master/train_infer_python.txt
+===========================train_params===========================
+model_name:table_master
+python:python3.7
+gpu_list:0|0,1
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=17
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:./pretrain_models/table_structure_tablemaster_train/best_accuracy
+train_model_name:latest
+train_infer_img_dir:./ppstructure/docs/table/table.jpg
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/table_master/table_master.yml -o Global.print_batch_step=10
+pact_train:null
+fpgm_train:null
+distill_train:null
+null:null
+null:null
+##
+===========================eval_params=========================== 
+eval:null
+null:null
+##
+===========================infer_params===========================
+Global.save_inference_dir:./output/
+Global.checkpoints:
+norm_export:tools/export_model.py -c test_tipc/configs/table_master/table_master.yml -o 
+quant_export:
+fpgm_export: 
+distill_export:null
+export1:null
+export2:null
+##
+infer_model:null
+infer_export:null
+infer_quant:False
+inference:ppstructure/table/predict_structure.py  --table_char_dict_path=./ppocr/utils/dict/table_master_structure_dict.txt --image_dir=./ppstructure/docs/table/table.jpg --output ./output/table --table_algorithm=TableMaster --table_max_len=480 
+--use_gpu:True|False
+--enable_mkldnn:False
+--cpu_threads:6
+--rec_batch_num:1
+--use_tensorrt:False
+--precision:fp32
+--table_model_dir:
+--image_dir:./ppstructure/docs/table/table.jpg
+null:null
+--benchmark:False
+null:null
+===========================infer_benchmark_params==========================
+random_infer_input:[{float32,[3,480,480]}]
--- a/test_tipc/prepare.sh
+++ b/test_tipc/prepare.sh
@@ -62,6 +62,10 @@ if [ ${MODE} = "lite_train_lite_infer" ];then
    if [[ ${model_name} =~ "det_r50_db++" ]];then
        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/dygraph_v2.1/en_det/ResNet50_dcn_asf_synthtext_pretrained.pdparams --no-check-certificate
    fi
+    if [ ${model_name} == "table_master" ];then
+        wget -nc -P ./pretrain_models/ https://paddleocr.bj.bcebos.com/ppstructure/models/tablemaster/table_structure_tablemaster_train.tar --no-check-certificate
+        cd ./pretrain_models/ && tar xf table_structure_tablemaster_train.tar  && cd ../
+    fi
    cd ./pretrain_models/ && tar xf det_mv3_db_v2.0_train.tar && cd ../
    rm -rf ./train_data/icdar2015
    rm -rf ./train_data/ic15_data

--- a/test_tipc/readme.md
+++ b/test_tipc/readme.md
@@ -54,6 +54,7 @@
 | NRTR   |rec_mtb_nrtr                   | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
 | SAR    |rec_r31_sar                    | 识别  | 支持 | 多机多卡 <br> 混合精度 | - | - |
 | PGNet  |rec_r34_vd_none_none_ctc_v2.0  | 端到端| 支持 | 多机多卡 <br> 混合精度 | - | - |
+| TableMaster  |table_structure_tablemaster_train  | 表格识别| 支持 | 多机多卡 <br> 混合精度 | - | - |