Update fine-tuned det (hmean 0.916) + rec + dict 233 chars + ViLM corpus

.gitattributes

@@ -35,3 +35,6 @@ saved_model/**/* filter=lfs diff=lfs merge=lfs -text
 *tfevents* filter=lfs diff=lfs merge=lfs -text
 models/inference_det/inference.pdiparams filter=lfs diff=lfs merge=lfs -text
 models/inference_rec/inference.pdiparams filter=lfs diff=lfs merge=lfs -text
+lm_data/vi_corpus.txt filter=lfs diff=lfs merge=lfs -text
+models/det_best_accuracy.pdparams filter=lfs diff=lfs merge=lfs -text
+models/rec_best_accuracy.pdparams filter=lfs diff=lfs merge=lfs -text

dict/vi_vietnam.txt

@@ -0,0 +1,233 @@
+!
+"
+#
+$
+%
+&
+'
+(
+)
+*
++
+,
+-
+.
+/
+0
+1
+2
+3
+4
+5
+6
+7
+8
+9
+:
+;
+<
+=
+>
+?
+A
+B
+C
+D
+E
+F
+G
+H
+I
+J
+K
+L
+M
+N
+O
+P
+Q
+R
+S
+T
+U
+V
+W
+X
+Y
+[
+\
+]
+a
+b
+c
+d
+e
+f
+g
+h
+i
+j
+k
+l
+m
+n
+o
+p
+q
+r
+s
+t
+u
+v
+w
+x
+y
+z
+{
+|
+}
+«
+²
+¹
+»
+À
+Á
+Â
+Ã
+É
+Ê
+Ì
+Í
+Ò
+Ó
+Ô
+Ù
+Ú
+Ý
+à
+á
+â
+ã
+ä
+ç
+è
+é
+ê
+ì
+í
+ò
+ó
+ô
+õ
+ù
+ú
+ü
+ý
+ā
+Ă
+ă
+Đ
+đ
+ē
+ě
+Ĩ
+ĩ
+ī
+ō
+Ũ
+ũ
+ū
+Ơ
+ơ
+Ư
+ư
+ǎ
+ǒ
+ǔ
+Ạ
+ạ
+Ả
+ả
+Ấ
+ấ
+Ầ
+ầ
+Ẩ
+ẩ
+ẫ
+Ậ
+ậ
+Ắ
+ắ
+ằ
+ẳ
+ẵ
+Ặ
+ặ
+ẹ
+Ẻ
+ẻ
+ẽ
+Ế
+ế
+Ề
+ề
+Ể
+ể
+Ễ
+ễ
+Ệ
+ệ
+Ỉ
+ỉ
+Ị
+ị
+Ọ
+ọ
+Ỏ
+ỏ
+Ố
+ố
+Ồ
+ồ
+Ổ
+ổ
+ỗ
+Ộ
+ộ
+Ớ
+ớ
+Ờ
+ờ
+Ở
+ở
+ỡ
+Ợ
+ợ
+Ụ
+ụ
+Ủ
+ủ
+Ứ
+ứ
+Ừ
+ừ
+Ử
+ử
+Ữ
+ữ
+Ự
+ự
+Ỳ
+ỳ
+ỵ
+Ỷ
+ỷ
+ỹ
+–
+—
+’
+“
+”
+•
+…

lm_data/vi_corpus.txt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:419af78bc5158caadaa046a6f1f82ed2c6b6a938219755b1c0354714d6e5381a
+size 64558088

models/config_det.yml

@@ -0,0 +1,139 @@
+
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 30
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: /kaggle/working/output/det_v5m
+  save_epoch_step: 1000
+  eval_batch_step: [2000, 2000]
+  cal_metric_during_train: false
+  pretrained_model: /kaggle/working/PP-OCRv5_mobile_det_pretrained.pdparams
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  save_res_path: /kaggle/working/predicts_db.txt
+  distributed: true
+  model_name: PP-OCRv5_mobile_det
+
+Architecture:
+  model_type: det
+  algorithm: DB
+  Transform: null
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.75
+    det: True
+  Neck:
+    name: RSEFPN
+    out_channels: 96
+    shortcut: True
+  Head:
+    name: DBHead
+    k: 50
+
+Loss:
+  name: DBLoss
+  balance_loss: true
+  main_loss_type: DiceLoss
+  alpha: 5
+  beta: 10
+  ohem_ratio: 3
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0005
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 5.0e-05
+
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: /kaggle/input/datasets/sirimiriiii13/vi-det-full
+    label_file_list:
+      - /kaggle/input/datasets/sirimiriiii13/vi-det-full/train.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: false
+      - DetLabelEncode: null
+      - IaaAugment:
+          augmenter_args:
+            - type: Affine
+              args:
+                rotate: [-10, 10]
+            - type: Resize
+              args:
+                size: [0.5, 3]
+      - EastRandomCropData:
+          size: [640, 640]
+          max_tries: 50
+          keep_ratio: true
+      - MakeBorderMap:
+          shrink_ratio: 0.4
+          thresh_min: 0.3
+          thresh_max: 0.7
+      - MakeShrinkMap:
+          shrink_ratio: 0.4
+          min_text_size: 3
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, threshold_map, threshold_mask, shrink_map, shrink_mask]
+  loader:
+    shuffle: true
+    drop_last: true
+    batch_size_per_card: 8
+    num_workers: 4
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: /kaggle/input/datasets/sirimiriiii13/vi-det-full
+    label_file_list:
+      - /kaggle/input/datasets/sirimiriiii13/vi-det-full/val.txt
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: false
+      - DetLabelEncode: null
+      - DetResizeForTest:
+          limit_side_len: 1280
+          limit_type: max
+      - NormalizeImage:
+          scale: 1./255.
+          mean: [0.485, 0.456, 0.406]
+          std: [0.229, 0.224, 0.225]
+          order: hwc
+      - ToCHWImage: null
+      - KeepKeys:
+          keep_keys: [image, shape, polys, ignore_tags]
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 1
+    num_workers: 2

models/det_best_accuracy.pdparams

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:213266d0a626cdc14f7e10e6436ac6d02e14f6060189731c254193e98bd24af7
+size 14381328

models/inference_det/inference.pdiparams

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:7d16739f6f7e9db25b7eea78425ee1eeb46b3a7eda94e9e363baa1200259bfa7
-size 2377917
+oid sha256:ca8be3da54703bd891bbd899d27f57429a4a76aad5a21c0c56e2a3f785cc841d
+size 4692097

models/inference_det/inference.yml

@@ -1,5 +1,5 @@
 Global:
-  model_name: PP-OCRv5_server_det
+  model_name: PP-OCRv5_mobile_det
 Hpi:
   backend_configs:
     paddle_infer:
@@ -19,12 +19,6 @@ Hpi:
           - 4000
     tensorrt:
       dynamic_shapes: *id001
-PostProcess:
-  box_thresh: 0.6
-  max_candidates: 1000
-  name: DBPostProcess
-  thresh: 0.3
-  unclip_ratio: 1.5
 PreProcess:
   transform_ops:
   - DecodeImage:
@@ -32,8 +26,7 @@ PreProcess:
       img_mode: BGR
   - DetLabelEncode: null
   - DetResizeForTest:
-      limit_side_len: 1280
-      limit_type: max
+      resize_long: 960
   - NormalizeImage:
       mean:
       - 0.485
@@ -52,3 +45,9 @@ PreProcess:
       - shape
       - polys
       - ignore_tags
+PostProcess:
+  name: DBPostProcess
+  thresh: 0.3
+  box_thresh: 0.6
+  max_candidates: 1000
+  unclip_ratio: 1.5

models/inference_rec/inference.pdiparams

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b27d4b43fbec4d719ca93937855b4e21fe79edf38d77c9cdf95ffe931ce6612c
-size 7627599
+oid sha256:1988f75d1a0282a59fc9fdd99f0eee70f4b7c5b695fd28b6d22fa9974ff772a3
+size 7674063

models/inference_rec/inference.yml

@@ -3,8 +3,8 @@ Global:
 Hpi:
   backend_configs:
     paddle_infer:
-      trt_dynamic_shapes: &id001
-        x:
+      trt_dynamic_shapes:
+        x: &id001
         - - 1
           - 3
           - 48
@@ -18,15 +18,20 @@ Hpi:
           - 48
           - 3200
     tensorrt:
-      dynamic_shapes: *id001
+      dynamic_shapes:
+        x: *id001
 PostProcess:
   character_dict:
   - '!'
   - '"'
+  - '#'
+  - $
   - '%'
+  - '&'
   - ''''
   - (
   - )
+  - '*'
   - +
   - ','
   - '-'
@@ -44,16 +49,20 @@ PostProcess:
   - '9'
   - ':'
   - ;
+  - <
   - '='
+  - '>'
   - '?'
   - A
   - B
   - C
   - D
   - E
+  - F
   - G
   - H
   - I
+  - J
   - K
   - L
   - M
@@ -66,16 +75,22 @@ PostProcess:
   - T
   - U
   - V
+  - W
   - X
   - Y
+  - '['
+  - \
+  - ']'
   - a
   - b
   - c
   - d
   - e
+  - f
   - g
   - h
   - i
+  - j
   - k
   - l
   - m
@@ -88,12 +103,37 @@ PostProcess:
   - t
   - u
   - v
+  - w
   - x
   - y
+  - z
+  - '{'
+  - '|'
+  - '}'
+  - «
+  - ²
+  - ¹
+  - »
+  - À
+  - Á
+  - Â
+  - Ã
+  - É
+  - Ê
+  - Ì
+  - Í
+  - Ò
+  - Ó
+  - Ô
+  - Ù
+  - Ú
+  - Ý
   - à
   - á
   - â
   - ã
+  - ä
+  - ç
   - è
   - é
   - ê
@@ -105,59 +145,116 @@ PostProcess:
   - õ
   - ù
   - ú
+  - ü
   - ý
+  - ā
+  - Ă
   - ă
   - Đ
   - đ
+  - ē
+  - ě
+  - Ĩ
   - ĩ
+  - ī
+  - ō
+  - Ũ
   - ũ
+  - ū
+  - Ơ
   - ơ
+  - Ư
   - ư
+  - ǎ
+  - ǒ
+  - ǔ
+  - Ạ
   - ạ
+  - Ả
   - ả
+  - Ấ
   - ấ
+  - Ầ
   - ầ
+  - Ẩ
   - ẩ
   - ẫ
+  - Ậ
   - ậ
+  - Ắ
   - ắ
   - ằ
   - ẳ
   - ẵ
+  - Ặ
   - ặ
   - ẹ
+  - Ẻ
   - ẻ
   - ẽ
+  - Ế
   - ế
+  - Ề
   - ề
+  - Ể
   - ể
+  - Ễ
   - ễ
+  - Ệ
   - ệ
+  - Ỉ
   - ỉ
+  - Ị
   - ị
+  - Ọ
   - ọ
+  - Ỏ
   - ỏ
+  - Ố
   - ố
+  - Ồ
   - ồ
+  - Ổ
   - ổ
   - ỗ
+  - Ộ
   - ộ
+  - Ớ
   - ớ
+  - Ờ
   - ờ
+  - Ở
   - ở
   - ỡ
+  - Ợ
   - ợ
+  - Ụ
   - ụ
+  - Ủ
   - ủ
+  - Ứ
   - ứ
+  - Ừ
   - ừ
+  - Ử
   - ử
+  - Ữ
   - ữ
+  - Ự
   - ự
+  - Ỳ
   - ỳ
   - ỵ
+  - Ỷ
   - ỷ
   - ỹ
+  - –
+  - —
+  - ’
+  - “
+  - ”
+  - •
+  - …
   name: CTCLabelDecode
 PreProcess:
   transform_ops:
@@ -166,13 +263,13 @@ PreProcess:
       img_mode: BGR
   - MultiLabelEncode:
       gtc_encode: NRTRLabelEncode
-      max_text_length: 96
+      max_text_length: 80
   - RecResizeImg:
       eval_mode: true
       image_shape:
       - 3
       - 48
-      - 640
+      - 960
   - KeepKeys:
       keep_keys:
       - image

models/rec_best_accuracy.pdparams

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1a788895640ff00dd04d5ae22acbd9a7402d9f4f3306a06960afadded88d8750
+size 69348604

models/vi_PP-OCRv5_mobile_rec.yml

@@ -0,0 +1,123 @@
+
+Global:
+  debug: false
+  use_gpu: true
+  epoch_num: 12
+  log_smooth_window: 20
+  print_batch_step: 20
+  save_model_dir: /kaggle/working/output/vi_ppocrv5
+  save_epoch_step: 1000
+  eval_batch_step: [0, 2000]
+  cal_metric_during_train: true
+  pretrained_model: /kaggle/input/datasets/sirimiriiii13/rec-ckpt-v14/output/vi_ppocrv5/best_accuracy
+  checkpoints: null
+  save_inference_dir: null
+  use_visualdl: false
+  infer_img: null
+  character_dict_path: /kaggle/input/datasets/sirimiriiii13/vi-rec-100k/vi_dict.txt
+  max_text_length: 80
+  infer_mode: false
+  use_space_char: true
+  distributed: true
+  save_res_path: /kaggle/working/rec_predicts.txt
+  d2s_train_image_shape: [3, 48, 960]
+
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+    name: Cosine
+    learning_rate: 0.0002
+    warmup_epoch: 2
+  regularizer:
+    name: L2
+    factor: 3.0e-05
+
+Architecture:
+  model_type: rec
+  algorithm: SVTR_LCNet
+  Transform: null
+  Backbone:
+    name: PPLCNetV3
+    scale: 0.95
+  Head:
+    name: MultiHead
+    head_list:
+      - CTCHead:
+          Neck:
+            name: svtr
+            dims: 120
+            depth: 2
+            hidden_dims: 120
+            kernel_size: [1, 3]
+            use_guide: true
+          Head:
+            fc_decay: 1.0e-05
+      - NRTRHead:
+          nrtr_dim: 384
+          max_text_length: 80
+
+Loss:
+  name: MultiLoss
+  loss_config_list:
+    - CTCLoss:
+    - NRTRLoss:
+
+PostProcess:
+  name: CTCLabelDecode
+
+Metric:
+  name: RecMetric
+  main_indicator: acc
+
+Train:
+  dataset:
+    name: MultiScaleDataSet
+    ds_width: false
+    data_dir: /kaggle/input/datasets/sirimiriiii13/vi-rec-100k
+    label_file_list:
+      - /kaggle/input/datasets/sirimiriiii13/vi-rec-100k/rec_train.txt
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: false
+      - RecAug: null
+      - MultiLabelEncode:
+          gtc_encode: NRTRLabelEncode
+      - KeepKeys:
+          keep_keys: [image, label_ctc, label_gtc, length, valid_ratio]
+  sampler:
+    name: MultiScaleSampler
+    scales: [[960, 32], [960, 48], [960, 64]]
+    first_bs: 24
+    fix_bs: false
+    divided_factor: [8, 16]
+    is_training: true
+  loader:
+    shuffle: true
+    batch_size_per_card: 24
+    drop_last: true
+    num_workers: 2
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: /kaggle/input/datasets/sirimiriiii13/vi-rec-100k
+    label_file_list:
+      - /kaggle/input/datasets/sirimiriiii13/vi-rec-100k/rec_val.txt
+    transforms:
+      - DecodeImage:
+          img_mode: BGR
+          channel_first: false
+      - MultiLabelEncode:
+          gtc_encode: NRTRLabelEncode
+      - RecResizeImg:
+          image_shape: [3, 48, 960]
+      - KeepKeys:
+          keep_keys: [image, label_ctc, label_gtc, length, valid_ratio]
+  loader:
+    shuffle: false
+    drop_last: false
+    batch_size_per_card: 32
+    num_workers: 2