Merge branch 'dygraph' into test_v11

ppocr/postprocess/east_postprocess.py

@@ -20,6 +20,7 @@ import numpy as np
 from .locality_aware_nms import nms_locality
 import cv2
 import paddle
+import lanms
 
 import os
 import sys
@@ -29,6 +30,7 @@ class EASTPostProcess(object):
     """
     The post process for EAST.
     """
+
     def __init__(self,
                  score_thresh=0.8,
                  cover_thresh=0.1,
@@ -38,11 +40,6 @@ class EASTPostProcess(object):
         self.score_thresh = score_thresh
         self.cover_thresh = cover_thresh
         self.nms_thresh = nms_thresh
-        
-        # c++ la-nms is faster, but only support python 3.5
-        self.is_python35 = False
-        if sys.version_info.major == 3 and sys.version_info.minor == 5:
-            self.is_python35 = True
 
     def restore_rectangle_quad(self, origin, geometry):
         """
@@ -79,11 +76,8 @@ class EASTPostProcess(object):
         boxes = np.zeros((text_box_restored.shape[0], 9), dtype=np.float32)
         boxes[:, :8] = text_box_restored.reshape((-1, 8))
         boxes[:, 8] = score_map[xy_text[:, 0], xy_text[:, 1]]
-        if self.is_python35:
-            import lanms
-            boxes = lanms.merge_quadrangle_n9(boxes, nms_thresh)
-        else:
-            boxes = nms_locality(boxes.astype(np.float64), nms_thresh)
+        boxes = lanms.merge_quadrangle_n9(boxes, nms_thresh)
+        # boxes = nms_locality(boxes.astype(np.float64), nms_thresh)
         if boxes.shape[0] == 0:
             return []
         # Here we filter some low score boxes by the average score map, 
@@ -139,4 +133,4 @@ class EASTPostProcess(object):
                         continue
                     boxes_norm.append(box)
             dt_boxes_list.append({'points': np.array(boxes_norm)})
-        return dt_boxes_list
\ No newline at end of file
+        return dt_boxes_list

ppocr/utils/save_load.py

@@ -54,11 +54,28 @@ def load_model(config, model, optimizer=None):
     pretrained_model = global_config.get('pretrained_model')
     best_model_dict = {}
     if checkpoints:
-        if checkpoints.endswith('pdparams'):
+        if checkpoints.endswith('.pdparams'):
             checkpoints = checkpoints.replace('.pdparams', '')
-        assert os.path.exists(checkpoints + ".pdopt"), \
-            f"The {checkpoints}.pdopt does not exists!"
-        load_pretrained_params(model, checkpoints)
+        assert os.path.exists(checkpoints + ".pdparams"), \
+            "The {}.pdparams does not exists!".format(checkpoints)
+
+        # load params from trained model
+        params = paddle.load(checkpoints + '.pdparams')
+        state_dict = model.state_dict()
+        new_state_dict = {}
+        for key, value in state_dict.items():
+            if key not in params:
+                logger.warning("{} not in loaded params {} !".format(
+                    key, params.keys()))
+            pre_value = params[key]
+            if list(value.shape) == list(pre_value.shape):
+                new_state_dict[key] = pre_value
+            else:
+                logger.warning(
+                    "The shape of model params {} {} not matched with loaded params shape {} !".
+                    format(key, value.shape, pre_value.shape))
+        model.set_state_dict(new_state_dict)
+
         optim_dict = paddle.load(checkpoints + '.pdopt')
         if optimizer is not None:
             optimizer.set_state_dict(optim_dict)
@@ -80,10 +97,10 @@ def load_model(config, model, optimizer=None):
 
 def load_pretrained_params(model, path):
     logger = get_logger()
-    if path.endswith('pdparams'):
+    if path.endswith('.pdparams'):
         path = path.replace('.pdparams', '')
     assert os.path.exists(path + ".pdparams"), \
-        f"The {path}.pdparams does not exists!"
+        "The {}.pdparams does not exists!".format(path)
 
     params = paddle.load(path + '.pdparams')
     state_dict = model.state_dict()
@@ -92,11 +109,11 @@ def load_pretrained_params(model, path):
         if list(state_dict[k1].shape) == list(params[k2].shape):
             new_state_dict[k1] = params[k2]
         else:
-            logger.info(
-                f"The shape of model params {k1} {state_dict[k1].shape} not matched with loaded params {k2} {params[k2].shape} !"
-            )
+            logger.warning(
+                "The shape of model params {} {} not matched with loaded params {} {} !".
+                format(k1, state_dict[k1].shape, k2, params[k2].shape))
     model.set_state_dict(new_state_dict)
-    logger.info(f"load pretrain successful from {path}")
+    logger.info("load pretrain successful from {}".format(path))
     return model
 
 

requirements.txt

@@ -12,4 +12,5 @@ cython
 lxml
 premailer
 openpyxl
-fasttext==0.9.1
\ No newline at end of file
+fasttext==0.9.1
+lanms-nova
\ No newline at end of file

test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt

+===========================lite_params===========================
+inference:./ocr_db_crnn det
+runtime_device:ARM_GPU_OPENCL
+det_infer_model:ch_PP-OCRv2_det_infer|ch_PP-OCRv2_det_slim_quant_infer
+null:null
+null:null
+--cpu_threads:1|4
+--det_batch_size:1
+null:null
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+null:null
+--benchmark:True

test_tipc/configs/ch_ppocr_mobile_v2.0/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt

+===========================lite_params===========================
+inference:./ocr_db_crnn system
+runtime_device:ARM_CPU
+det_infer_model:ch_ppocr_mobile_v2.0_det_infer|ch_ppocr_db_mobile_v2.0_det_quant_infer
+rec_infer_model:ch_ppocr_mobile_v2.0_rec_infer|ch_ppocr_mobile_v2.0_rec_slim_infer
+cls_infer_model:ch_ppocr_mobile_v2.0_cls_infer|ch_ppocr_mobile_v2.0_cls_slim_infer
+--cpu_threads:1|4
+--det_batch_size:1
+--rec_batch_size:1
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+--rec_dict_dir:./ppocr_keys_v1.txt
+--benchmark:True

test_tipc/configs/ch_ppocr_mobile_v2.0/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt

+===========================lite_params===========================
+inference:./ocr_db_crnn system
+runtime_device:ARM_GPU_OPENCL
+det_infer_model:ch_ppocr_mobile_v2.0_det_infer|ch_ppocr_db_mobile_v2.0_det_quant_infer
+rec_infer_model:ch_ppocr_mobile_v2.0_rec_infer|ch_ppocr_mobile_v2.0_rec_slim_infer
+cls_infer_model:ch_ppocr_mobile_v2.0_cls_infer|ch_ppocr_mobile_v2.0_cls_slim_infer
+--cpu_threads:1|4
+--det_batch_size:1
+--rec_batch_size:1
+--image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
+--config_dir:./config.txt
+--rec_dict_dir:./ppocr_keys_v1.txt
+--benchmark:True

test_tipc/configs/ch_ppocr_mobile_v2.0_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt

 ===========================lite_params===========================
 inference:./ocr_db_crnn det
-infer_model:ch_PP-OCRv2_det_infer|ch_PP-OCRv2_det_slim_quant_infer
 runtime_device:ARM_CPU
+det_infer_model:ch_ppocr_mobile_v2.0_det_infer|ch_ppocr_db_mobile_v2.0_det_quant_infer
+null:null
+null:null
 --cpu_threads:1|4
 --det_batch_size:1
---rec_batch_size:1
---system_batch_size:1
+null:null
 --image_dir:./test_data/icdar2015_lite/text_localization/ch4_test_images/
 --config_dir:./config.txt
---rec_dict_dir:./ppocr_keys_v1.txt
+null:null
 --benchmark:True

test_tipc/configs/ch_ppocr_mobile_v2.0_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt

 ===========================lite_params===========================
 inference:./ocr_db_crnn det
 runtime_device:ARM_GPU_OPENCL
-det_infer_model:ch_PP-OCRv2_det_infer|ch_PP-OCRv2_det_slim_quant_infer
+det_infer_model:ch_ppocr_mobile_v2.0_det_infer|ch_ppocr_db_mobile_v2.0_det_quant_infer
 null:null
 null:null
 --cpu_threads:1|4

test_tipc/configs/det_mv3_east_v2.0/det_mv3_east.yml

+Global:
+  use_gpu: true
+  epoch_num: 10000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/east_mv3/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: ./pretrain_models/MobileNetV3_large_x0_5_pretrained
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  save_res_path: ./output/det_east/predicts_east.txt
+
+Architecture:
+  model_type: det
+  algorithm: EAST
+  Transform:
+  Backbone:
+    name: MobileNetV3
+    scale: 0.5
+    model_name: large
+  Neck:
+    name: EASTFPN
+    model_name: small
+  Head:
+    name: EASTHead
+    model_name: small
+
+Loss:
+  name: EASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: EASTPostProcess
+  score_thresh: 0.8
+  cover_thresh: 0.1
+  nms_thresh: 0.2
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - EASTProcessTrain:
+          image_shape: [512, 512]
+          background_ratio: 0.125
+          min_crop_side_ratio: 0.1
+          min_text_size: 10
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'geo_map', 'training_mask'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 16
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 2400
+          limit_type: max
+      - 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', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
\ No newline at end of file

test_tipc/configs/det_mv3_east_v2.0/train_infer_python.txt

+===========================train_params===========================
+model_name:det_mv3_east_v2.0
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_mv3_east_v2.0/det_mv3_east.yml -o 
+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.pretrained_model:
+norm_export:tools/export_model.py -c test_tipc/configs/det_mv3_east_v2.0/det_mv3_east.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_mv3_east/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/cconfigs/det_mv3_east_v2.0/det_mv3_east.yml  -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--det_algorithm:EAST

test_tipc/configs/det_r50_vd_east_v2.0/det_r50_vd_east.yml

+Global:
+  use_gpu: true
+  epoch_num: 10000
+  log_smooth_window: 20
+  print_batch_step: 2
+  save_model_dir: ./output/east_r50_vd/
+  save_epoch_step: 1000
+  # evaluation is run every 5000 iterations after the 4000th iteration
+  eval_batch_step: [4000, 5000]
+  cal_metric_during_train: False
+  pretrained_model: 
+  checkpoints: 
+  save_inference_dir:
+  use_visualdl: False
+  infer_img: 
+  save_res_path: ./output/det_east/predicts_east.txt
+
+Architecture:
+  model_type: det
+  algorithm: EAST
+  Transform:
+  Backbone:
+    name: ResNet
+    layers: 50
+  Neck:
+    name: EASTFPN
+    model_name: large
+  Head:
+    name: EASTHead
+    model_name: large
+
+Loss:
+  name: EASTLoss
+  
+Optimizer:
+  name: Adam
+  beta1: 0.9
+  beta2: 0.999
+  lr:
+  #  name: Cosine
+    learning_rate: 0.001
+  #  warmup_epoch: 0
+  regularizer:
+    name: 'L2'
+    factor: 0
+
+PostProcess:
+  name: EASTPostProcess
+  score_thresh: 0.8
+  cover_thresh: 0.1
+  nms_thresh: 0.2
+
+Metric:
+  name: DetMetric
+  main_indicator: hmean
+
+Train:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/train_icdar2015_label.txt
+    ratio_list: [1.0]
+    transforms:
+      - DecodeImage: # load image
+          img_mode: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - EASTProcessTrain:
+          image_shape: [512, 512]
+          background_ratio: 0.125
+          min_crop_side_ratio: 0.1
+          min_text_size: 10
+      - KeepKeys:
+          keep_keys: ['image', 'score_map', 'geo_map', 'training_mask'] # dataloader will return list in this order
+  loader:
+    shuffle: True
+    drop_last: False
+    batch_size_per_card: 8
+    num_workers: 8
+
+Eval:
+  dataset:
+    name: SimpleDataSet
+    data_dir: ./train_data/icdar2015/text_localization/
+    label_file_list:
+      - ./train_data/icdar2015/text_localization/test_icdar2015_label.txt
+    transforms:
+      - DecodeImage: # load image
+          Fa: BGR
+          channel_first: False
+      - DetLabelEncode: # Class handling label
+      - DetResizeForTest:
+          limit_side_len: 2400
+          limit_type: max
+      - 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', 'ignore_tags']
+  loader:
+    shuffle: False
+    drop_last: False
+    batch_size_per_card: 1 # must be 1
+    num_workers: 2
\ No newline at end of file

test_tipc/configs/det_r50_vd_east_v2.0/train_infer_python.txt

+===========================train_params===========================
+model_name:det_r50_vd_east_v2.0
+python:python3.7
+gpu_list:0
+Global.use_gpu:True|True
+Global.auto_cast:fp32
+Global.epoch_num:lite_train_lite_infer=1|whole_train_whole_infer=500
+Global.save_model_dir:./output/
+Train.loader.batch_size_per_card:lite_train_lite_infer=2|whole_train_whole_infer=4
+Global.pretrained_model:null
+train_model_name:latest
+train_infer_img_dir:./train_data/icdar2015/text_localization/ch4_test_images/
+null:null
+##
+trainer:norm_train
+norm_train:tools/train.py -c test_tipc/configs/det_r50_vd_east_v2.0/det_r50_vd_east.yml -o 
+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.pretrained_model:
+norm_export:tools/export_model.py -c test_tipc/configs/det_r50_vd_east_v2.0/det_r50_vd_east.yml -o 
+quant_export:null 
+fpgm_export:null
+distill_export:null
+export1:null
+export2:null
+##
+train_model:./inference/det_mv3_east/best_accuracy
+infer_export:tools/export_model.py -c test_tipc/cconfigs/det_r50_vd_east_v2.0/det_r50_vd_east.yml  -o
+infer_quant:False
+inference:tools/infer/predict_det.py
+--use_gpu:True|False
+--enable_mkldnn:True|False
+--cpu_threads:1|6
+--rec_batch_num:1
+--use_tensorrt:False|True
+--precision:fp32|fp16|int8
+--det_model_dir:
+--image_dir:./inference/ch_det_data_50/all-sum-510/
+--save_log_path:null
+--benchmark:True
+--det_algorithm:EAST

test_tipc/docs/test_lite_arm_cpp.md

@@ -16,7 +16,7 @@ Lite\_arm\_cpp预测功能测试的主程序为`test_lite_arm_cpp.sh`，可以
 
 | 模型类型 | batch-size | threads | predictor数量 | 预测库来源 | 测试硬件 |
 |  :----:   |  :----:  | :----:  |  :----:  |  :----:  |  :----:  |
-| 正常模型/量化模型 | 1 | 1/4 |  单/多 | 下载方式 | ARM\_CPU/ARM\_GPU_OPENCL |
+| 正常模型/量化模型 | 1 | 1/4 |  单/多 | 下载方式/编译方式 | ARM\_CPU/ARM\_GPU_OPENCL |
 
 
 ## 2. 测试流程
@@ -30,8 +30,11 @@ Lite\_arm\_cpp预测功能测试的主程序为`test_lite_arm_cpp.sh`，可以
 
 ```shell
 
-# 数据和模型准备
-bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ppocr_det_mobile/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
+# 数据、模型、Paddle-Lite预测库准备
+#预测库为下载方式
+bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt download
+#预测库为编译方式
+bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt compile
 
 # 手机端测试:
 bash test_lite_arm_cpp.sh model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
@@ -42,8 +45,11 @@ bash test_lite_arm_cpp.sh model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt
 
 ```shell
 
-# 数据和模型准备
-bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ppocr_det_mobile/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
+# 数据、模型、Paddle-Lite预测库准备
+#预测库下载方式
+bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt download
+#预测库编译方式
+bash test_tipc/prepare_lite_cpp.sh ./test_tipc/configs/ch_PP-OCRv2_det/model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt compile
 
 # 手机端测试:
 bash test_lite_arm_cpp.sh model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.txt
@@ -53,9 +59,7 @@ bash test_lite_arm_cpp.sh model_linux_gpu_normal_normal_lite_cpp_arm_gpu_opencl.
 
 **注意**：
 
-1.由于运行该项目需要bash等命令，传统的adb方式不能很好的安装。所以此处推荐通在手机上开启虚拟终端的方式连接电脑，连接方式可以参考[安卓手机termux连接电脑](./termux_for_android.md)。
-
-2.如果测试文本检测和识别完整的pipeline，在执行`prepare_lite_cpp.sh`时，配置文件需替换为`test_tipc/configs/ppocr_system_mobile/model_linux_gpu_normal_normal_lite_cpp_arm_cpu.txt`。在手机端测试阶段，配置文件同样修改为该文件。
+由于运行该项目需要bash等命令，传统的adb方式不能很好的安装。所以此处推荐通在手机上开启虚拟终端的方式连接电脑，连接方式可以参考[安卓手机termux连接电脑](./termux_for_android.md)。
 
 ### 2.2 运行结果
 

test_tipc/prepare_lite_cpp.sh

@@ -6,6 +6,7 @@ dataline=$(cat ${FILENAME})
 IFS=$'\n'
 lines=(${dataline})
 IFS=$'\n'
+paddlelite_library_source=$2
 
 inference_cmd=$(func_parser_value "${lines[1]}")
 DEVICE=$(func_parser_value "${lines[2]}")
@@ -13,40 +14,42 @@ det_lite_model_list=$(func_parser_value "${lines[3]}")
 rec_lite_model_list=$(func_parser_value "${lines[4]}")
 cls_lite_model_list=$(func_parser_value "${lines[5]}")
 
-if [[ $inference_cmd =~ "det" ]];then
+if [[ $inference_cmd =~ "det" ]]; then
     lite_model_list=${det_lite_model_list}
-elif [[ $inference_cmd =~ "rec" ]];then
+elif [[ $inference_cmd =~ "rec" ]]; then
     lite_model_list=(${rec_lite_model_list[*]} ${cls_lite_model_list[*]})
-elif [[ $inference_cmd =~ "system" ]];then
+elif [[ $inference_cmd =~ "system" ]]; then
     lite_model_list=(${det_lite_model_list[*]} ${rec_lite_model_list[*]} ${cls_lite_model_list[*]})
 else
     echo "inference_cmd is wrong, please check."
     exit 1
 fi
 
-if [ ${DEVICE} = "ARM_CPU" ];then
+if [ ${DEVICE} = "ARM_CPU" ]; then
     valid_targets="arm"
-    paddlelite_url="https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10-rc/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz"
+    paddlelite_library_url="https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10-rc/inference_lite_lib.android.armv8.gcc.c++_shared.with_extra.with_cv.tar.gz"
     end_index="66"
-elif [ ${DEVICE} = "ARM_GPU_OPENCL" ];then
+    compile_with_opencl="OFF"
+elif [ ${DEVICE} = "ARM_GPU_OPENCL" ]; then
     valid_targets="opencl"
-    paddlelite_url="https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10-rc/inference_lite_lib.armv8.clang.with_exception.with_extra.with_cv.opencl.tar.gz"
+    paddlelite_library_url="https://github.com/PaddlePaddle/Paddle-Lite/releases/download/v2.10-rc/inference_lite_lib.armv8.clang.with_exception.with_extra.with_cv.opencl.tar.gz"
     end_index="71"
+    compile_with_opencl="ON"
 else
-    echo "DEVICE only suport ARM_CPU, ARM_GPU_OPENCL."
+    echo "DEVICE only support ARM_CPU, ARM_GPU_OPENCL."
     exit 2    
 fi
 
-# prepare lite .nb model
+# prepare paddlelite model
 pip install paddlelite==2.10-rc
 current_dir=${PWD}
 IFS="|"
 model_path=./inference_models
 
 for model in ${lite_model_list[*]}; do
-    if [[ $model =~ "PP-OCRv2" ]];then
+    if [[ $model =~ "PP-OCRv2" ]]; then
         inference_model_url=https://paddleocr.bj.bcebos.com/PP-OCRv2/chinese/${model}.tar
-    elif [[ $model =~ "v2.0" ]];then
+    elif [[ $model =~ "v2.0" ]]; then
         inference_model_url=https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/${model}.tar
     else 
         echo "Model is wrong, please check."
@@ -63,31 +66,42 @@ done
 
 # prepare test data
 data_url=https://paddleocr.bj.bcebos.com/dygraph_v2.0/test/icdar2015_lite.tar
-model_path=./inference_models
-inference_model=${inference_model_url##*/}
 data_file=${data_url##*/}
-wget -nc  -P ./inference_models ${inference_model_url}
 wget -nc  -P ./test_data ${data_url}
-cd ./inference_models && tar -xf ${inference_model} && cd ../
 cd ./test_data && tar -xf ${data_file} && rm ${data_file} && cd ../
 
-# prepare lite env
-paddlelite_zipfile=$(echo $paddlelite_url | awk -F "/" '{print $NF}')
-paddlelite_file=${paddlelite_zipfile:0:${end_index}}
-wget ${paddlelite_url} && tar -xf ${paddlelite_zipfile}
-mkdir -p  ${paddlelite_file}/demo/cxx/ocr/test_lite
-cp -r ${model_path}/*_opt.nb test_data ${paddlelite_file}/demo/cxx/ocr/test_lite
-cp ppocr/utils/ppocr_keys_v1.txt deploy/lite/config.txt ${paddlelite_file}/demo/cxx/ocr/test_lite
-cp -r ./deploy/lite/* ${paddlelite_file}/demo/cxx/ocr/
-cp ${paddlelite_file}/cxx/lib/libpaddle_light_api_shared.so ${paddlelite_file}/demo/cxx/ocr/test_lite
-cp ${FILENAME} test_tipc/test_lite_arm_cpp.sh test_tipc/common_func.sh ${paddlelite_file}/demo/cxx/ocr/test_lite
-cd ${paddlelite_file}/demo/cxx/ocr/
+# prepare paddlelite predict library
+if [[ ${paddlelite_library_source} = "download" ]]; then
+    paddlelite_library_zipfile=$(echo $paddlelite_library_url | awk -F "/" '{print $NF}')
+    paddlelite_library_file=${paddlelite_library_zipfile:0:${end_index}}
+    wget ${paddlelite_library_url} && tar -xf ${paddlelite_library_zipfile}
+    cd ${paddlelite_library_zipfile}
+elif [[ ${paddlelite_library_source} = "compile" ]]; then
+    git clone -b release/v2.10 https://github.com/PaddlePaddle/Paddle-Lite.git
+    cd Paddle-Lite
+    ./lite/tools/build_android.sh  --arch=armv8  --with_cv=ON --with_extra=ON --toolchain=clang --with_opencl=${compile_with_opencl}
+    cd ../
+    cp -r Paddle-Lite/build.lite.android.armv8.clang/inference_lite_lib.android.armv8/ .
+    paddlelite_library_file=inference_lite_lib.android.armv8
+else
+    echo "paddlelite_library_source only support 'download' and 'compile'"
+    exit 3
+fi
+
+# organize the required files  
+mkdir -p  ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cp -r ${model_path}/*_opt.nb test_data ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cp ppocr/utils/ppocr_keys_v1.txt deploy/lite/config.txt ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cp -r ./deploy/lite/* ${paddlelite_library_file}/demo/cxx/ocr/
+cp ${paddlelite_library_file}/cxx/lib/libpaddle_light_api_shared.so ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cp ${FILENAME} test_tipc/test_lite_arm_cpp.sh test_tipc/common_func.sh ${paddlelite_library_file}/demo/cxx/ocr/test_lite
+cd ${paddlelite_library_file}/demo/cxx/ocr/
 git clone https://github.com/cuicheng01/AutoLog.git
 
-# make
+# compile and do some postprocess
 make -j
 sleep 1
 make -j
 cp ocr_db_crnn test_lite && cp test_lite/libpaddle_light_api_shared.so test_lite/libc++_shared.so
 tar -cf test_lite.tar ./test_lite && cp test_lite.tar ${current_dir} && cd ${current_dir}
-rm -rf ${paddlelite_file}* && rm -rf ${model_path}
+rm -rf ${paddlelite_library_file}* && rm -rf ${model_path}

tools/infer/utility.py

@@ -190,6 +190,7 @@ def create_predictor(args, mode, logger):
             config.enable_use_gpu(args.gpu_mem, 0)
             if args.use_tensorrt:
                 config.enable_tensorrt_engine(
+                    workspace_size=1 << 30,
                     precision_mode=precision,
                     max_batch_size=args.max_batch_size,
                     min_subgraph_size=args.min_subgraph_size)
@@ -310,10 +311,6 @@ def create_predictor(args, mode, logger):
 
 
 def get_infer_gpuid():
-    cmd = "nvidia-smi"
-    res = os.popen(cmd).readlines()
-    if len(res) == 0:
-        return None
     cmd = "env | grep CUDA_VISIBLE_DEVICES"
     env_cuda = os.popen(cmd).readlines()
     if len(env_cuda) == 0: