Add Korean LP dictionary and PGNet config

- dict/kr_lp_dict.txt: 67 chars covering 4 plate types (10 digits + 40 usage hangul + 17 region hangul, dedup) - configs/kr_lp_pgnet.yml: PGNet config tuned for Korean LP (pad_num=67, max_text_length=10, valid_set=partvgg, infer_visual_type=CN) - setup_server.sh: symlink dict and config into PaddleOCR tree Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
1 month ago · 3e5f823dd5
parent 35c9f9cb5d
commit 3e5f823dd5
3 changed files with 196 additions and 2 deletions
--- a/configs/kr_lp_pgnet.yml
+++ b/configs/kr_lp_pgnet.yml
@ -0,0 +1,120 @@
+Global:
+  use_gpu: True
+  epoch_num: 200
+  log_smooth_window: 20
+  print_batch_step: 10
+  save_model_dir: ./output/kr_lp_pgnet/
+  save_epoch_step: 10
+  eval_batch_step: [0, 1000]
+  cal_metric_during_train: False
+  pretrained_model:
+  checkpoints:
+  save_inference_dir:
+  use_visualdl: False
+  infer_img:
+  infer_visual_type: CN
+  valid_set: partvgg
+  save_res_path: ./output/kr_lp_pgnet/predicts.txt
+  character_dict_path: ppocr/utils/kr_lp_dict.txt
+  character_type: CN
+  max_text_length: 10
+  max_text_nums: 5
+  tcl_len: 64
+
+Architecture:
+  model_type: e2e
+  algorithm: PGNet
+  Transform:
+  Backbone:
+    name: ResNet
+    layers: 50
+  Neck:
+    name: PGFPN
+  Head:
+    name: PGHead
+    character_dict_path: ppocr/utils/kr_lp_dict.txt
+
+Loss:
+  name: PGLoss
+  tcl_bs: 64
+  max_text_length: 10
+  max_text_nums: 5
+  pad_num: 67
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.001
+    warmup_epoch: 50
+  regularizer:
+    name: 'L2'
+    factor: 0.0001
+
+PostProcess:
+  name: PGPostProcess
+  score_thresh: 0.5
+  mode: fast
+  point_gather_mode: align
+
+Metric:
+  name: E2EMetric
+  mode: A
+  gt_mat_dir: ./train_data/kr_lp_synth/gt
+  character_dict_path: ppocr/utils/kr_lp_dict.txt
+  main_indicator: f_score_e2e
+
+Train:
+  dataset:
+    name: PGDataSet
+    data_dir: ./train_data/kr_lp_synth/train
+    label_file_list: [./train_data/kr_lp_synth/train/train.txt]
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - E2ELabelEncodeTrain:
+      - PGProcessTrain:
+          batch_size: 14
+          use_resize: True
+          use_random_crop: False
+          min_crop_size: 24
+          min_text_size: 4
+          max_text_size: 512
+          point_gather_mode: align
+      - KeepKeys:
+          keep_keys: ['images', 'tcl_maps', 'tcl_label_maps', 'border_maps', 'direction_maps', 'training_masks', 'label_list', 'pos_list', 'pos_mask']
+  loader:
+    shuffle: True
+    drop_last: True
+    batch_size_per_card: 14
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: PGDataSet
+    data_dir: ./train_data/kr_lp_synth/test
+    label_file_list: [./train_data/kr_lp_synth/test/test.txt]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: False
+      - E2ELabelEncodeTest:
+      - E2EResizeForTest:
+          max_side_len: 768
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: 'hwc'
+      - ToCHWImage:
+      - KeepKeys:
+          keep_keys: ['image', 'shape', 'polys', 'texts', 'ignore_tags', 'img_id']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1
+    num_workers: 2
--- a/dict/kr_lp_dict.txt
+++ b/dict/kr_lp_dict.txt
@ -0,0 +1,67 @@
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+가
+나
+다
+라
+마
+바
+사
+아
+자
+거
+너
+더
+러
+머
+버
+서
+어
+저
+고
+노
+도
+로
+모
+보
+소
+오
+조
+구
+누
+두
+루
+무
+부
+수
+우
+주
+하
+허
+호
+배
+울
+산
+대
+인
+천
+광
+전
+세
+종
+경
+기
+강
+원
+충
+북
+남
+제
--- a/scripts/setup_server.sh
+++ b/scripts/setup_server.sh
@ -5,7 +5,9 @@

 set -euo pipefail

-WORKSPACE="${WORKSPACE:-$HOME/workspace}"
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+KR_LP_DIR="$(cd "$SCRIPT_DIR/.." && pwd)"
+WORKSPACE="${WORKSPACE:-$(cd "$KR_LP_DIR/.." && pwd)}"
 PADDLEOCR_DIR="$WORKSPACE/PaddleOCR"
 PRETRAIN_DIR="$PADDLEOCR_DIR/pretrain_models"

@ -30,7 +32,12 @@ if [ ! -d "train_step1" ]; then
    rm train_step1.tar
 fi

-echo "[4/4] GPU/Paddle 동작 확인"
+echo "[4/5] kr_lp_dict 및 config를 PaddleOCR 트리에 symlink"
+ln -sf "$KR_LP_DIR/dict/kr_lp_dict.txt" "$PADDLEOCR_DIR/ppocr/utils/kr_lp_dict.txt"
+mkdir -p "$PADDLEOCR_DIR/configs/e2e"
+ln -sf "$KR_LP_DIR/configs/kr_lp_pgnet.yml" "$PADDLEOCR_DIR/configs/e2e/kr_lp_pgnet.yml"
+
+echo "[5/5] GPU/Paddle 동작 확인"
 python3 -c "import paddle; print('paddle:', paddle.__version__); print('CUDA available:', paddle.is_compiled_with_cuda()); print('GPU count:', paddle.device.cuda.device_count())"

 echo