first commit

2025-07-02 08:57:16 +03:00
commit 56532cc9a9
1901 changed files with 457695 additions and 0 deletions
--- a/configs/rec/LaTeX_OCR_rec.yaml
+++ b/configs/rec/LaTeX_OCR_rec.yaml
@@ -0,0 +1,131 @@
+Global:
+  model_name: LaTeX_OCR_rec # To use static model for inference.
+  use_gpu: True
+  epoch_num: 500
+  log_smooth_window: 20
+  print_batch_step: 100
+  save_model_dir: ./output/rec/latex_ocr/
+  save_epoch_step: 5
+  max_seq_len: 512
+  # evaluation is run every 60000 iterations (22 epoch)(batch_size = 56)
+  eval_batch_step: [0, 60000]
+  cal_metric_during_train: True
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: doc/datasets/pme_demo/0000013.png
+  infer_mode: False
+  use_space_char: False
+  rec_char_dict_path:  ppocr/utils/dict/latex_ocr_tokenizer.json
+  save_res_path: ./output/rec/predicts_latexocr.txt
+  d2s_train_image_shape: [1,256,256]
+  find_unused_parameters: True
+
+Optimizer:
+  name: AdamW
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Const
+    learning_rate: 0.0001
+
+Architecture:
+  model_type: rec
+  algorithm: LaTeXOCR
+  in_channels: 1
+  Transform:
+  Backbone:
+    name: HybridTransformer
+    img_size: [192, 672]
+    patch_size: 16
+    num_classes: 0
+    embed_dim: 256
+    depth: 4
+    num_heads: 8
+    input_channel: 1
+    is_predict: False
+    is_export: False
+  Head:
+    name: LaTeXOCRHead
+    pad_value: 0
+    is_export: False
+    decoder_args:
+      attn_on_attn: True
+      cross_attend: True
+      ff_glu: True
+      rel_pos_bias: False
+      use_scalenorm: False
+
+Loss:
+  name: LaTeXOCRLoss
+
+PostProcess:
+  name: LaTeXOCRDecode
+  rec_char_dict_path: ppocr/utils/dict/latex_ocr_tokenizer.json
+
+Metric:
+  name: LaTeXOCRMetric
+  main_indicator:  exp_rate
+  cal_bleu_score: True
+
+Train:
+  dataset:
+    name: LaTeXOCRDataSet
+    data_dir: ./train_data/LaTeXOCR/train
+    data: ./train_data/LaTeXOCR/latexocr_train.pkl
+    min_dimensions: [32, 32]
+    max_dimensions: [672, 192]
+    batch_size_per_pair: 56
+    keep_smaller_batches: False
+    transforms:
+      - DecodeImage:
+          channel_first: False
+      - MinMaxResize:
+          min_dimensions: [32, 32]
+          max_dimensions: [672, 192]        
+      - LatexTrainTransform:
+          bitmap_prob: .04
+      - NormalizeImage:
+          mean: [0.7931, 0.7931, 0.7931]
+          std: [0.1738, 0.1738, 0.1738]
+          order: 'hwc'
+      - LatexImageFormat:
+      - KeepKeys:
+          keep_keys: ['image']
+  loader:
+    shuffle: True
+    batch_size_per_card: 1
+    drop_last: False
+    num_workers: 0
+    collate_fn: LaTeXOCRCollator
+
+Eval:
+  dataset:
+    name: LaTeXOCRDataSet
+    data_dir: ./train_data/LaTeXOCR/val
+    data: ./train_data/LaTeXOCR/latexocr_val.pkl
+    min_dimensions: [32, 32]
+    max_dimensions: [672, 192]
+    batch_size_per_pair: 10
+    keep_smaller_batches: True
+    transforms:
+      - DecodeImage:
+          channel_first: False
+      - MinMaxResize:
+          min_dimensions: [32, 32]
+          max_dimensions: [672, 192]  
+      - LatexTestTransform:
+      - NormalizeImage:
+          mean: [0.7931, 0.7931, 0.7931]
+          std: [0.1738, 0.1738, 0.1738]
+          order: 'hwc'
+      - LatexImageFormat:
+      - KeepKeys:
+          keep_keys: ['image']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 0
+    collate_fn: LaTeXOCRCollator