import fastdeploy as fd import cv2 import os from subprocess import run def parse_arguments(): import argparse import ast parser = argparse.ArgumentParser() parser.add_argument( "--auto", required=True, help="Auto download, convert, compile and infer if True") parser.add_argument( "--det_model", required=True, help="Path of Detection model of PPOCR.") parser.add_argument( "--cls_model", required=True, help="Path of Classification model of PPOCR.") parser.add_argument( "--rec_model", required=True, help="Path of Recognization model of PPOCR.") parser.add_argument( "--rec_label_file", required=True, help="Path of Recognization label of PPOCR.") parser.add_argument( "--image", type=str, required=True, help="Path of test image file.") return parser.parse_args() def getPPOCRv3(): cmd_str_det = 'wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_infer.tar' tar_str_det = 'tar xvf ch_PP-OCRv3_det_infer.tar' cmd_str_cls = 'wget https://paddleocr.bj.bcebos.com/dygraph_v2.0/ch/ch_ppocr_mobile_v2.0_cls_infer.tar' tar_str_cls = 'tar xvf ch_ppocr_mobile_v2.0_cls_infer.tar' cmd_str_rec = 'wget https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_rec_infer.tar' tar_str_rec = 'tar xvf ch_PP-OCRv3_rec_infer.tar' cmd_str_img = 'wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/doc/imgs/12.jpg' cmd_str_label = 'wget https://gitee.com/paddlepaddle/PaddleOCR/raw/release/2.6/ppocr/utils/ppocr_keys_v1.txt' script_str = 'wget https://raw.githubusercontent.com/PaddlePaddle/Paddle2ONNX/develop/tools/paddle/paddle_infer_shape.py' if not os.path.exists('ch_PP-OCRv3_det_infer.tar'): print(cmd_str_det, tar_str_det) run(cmd_str_det, shell=True) run(tar_str_det, shell=True) if not os.path.exists('ch_ppocr_mobile_v2.0_cls_infer.tar'): print(cmd_str_cls, tar_str_cls) run(cmd_str_cls, shell=True) run(tar_str_cls, shell=True) if not os.path.exists('ch_PP-OCRv3_rec_infer.tar'): print(cmd_str_rec, tar_str_rec) run(cmd_str_rec, shell=True) run(tar_str_rec, shell=True) if not os.path.exists('12.jpg'): print(cmd_str_img) run(cmd_str_img, shell=True) if not os.path.exists('ppocr_keys_v1.txt'): print(cmd_str_label) run(cmd_str_label, shell=True) if not os.path.exists('paddle_infer_shape.py'): print(script_str) run(script_str, shell=True) def fix_input_shape(): fix_det_str = 'python paddle_infer_shape.py --model_dir ch_PP-OCRv3_det_infer \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --save_dir ch_PP-OCRv3_det_infer_fix \ --input_shape_dict="{\'x\':[1,3,960,608]}"' fix_rec_str = 'python paddle_infer_shape.py --model_dir ch_PP-OCRv3_rec_infer \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --save_dir ch_PP-OCRv3_rec_infer_fix \ --input_shape_dict="{\'x\':[1,3,48,320]}"' fix_cls_str = 'python paddle_infer_shape.py --model_dir ch_ppocr_mobile_v2.0_cls_infer \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --save_dir ch_PP-OCRv3_cls_infer_fix \ --input_shape_dict="{\'x\':[1,3,48,192]}"' print(fix_det_str) run(fix_det_str, shell=True) print(fix_rec_str) run(fix_rec_str, shell=True) print(fix_cls_str) run(fix_cls_str, shell=True) def paddle2onnx(): cmd_str_det = 'paddle2onnx --model_dir ch_PP-OCRv3_det_infer_fix \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --save_file ch_PP-OCRv3_det_infer.onnx \ --enable_dev_version True' cmd_str_cls = 'paddle2onnx --model_dir ch_PP-OCRv3_cls_infer_fix \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --save_file ch_PP-OCRv3_cls_infer.onnx \ --enable_dev_version True' cmd_str_rec = 'paddle2onnx --model_dir ch_PP-OCRv3_rec_infer_fix \ --model_filename inference.pdmodel \ --params_filename inference.pdiparams \ --save_file ch_PP-OCRv3_rec_infer.onnx \ --enable_dev_version True' print(cmd_str_det) run(cmd_str_det, shell=True) print(cmd_str_cls) run(cmd_str_cls, shell=True) print(cmd_str_rec) run(cmd_str_rec, shell=True) def mlir_prepare(): mlir_path = os.getenv("MODEL_ZOO_PATH") mlir_path = mlir_path[:-13] regression_path = os.path.join(mlir_path, 'regression') mv_str_list = ['mkdir ch_PP-OCRv3', 'cp -rf ' + os.path.join(regression_path, 'dataset/COCO2017/') + ' ./ch_PP-OCRv3', 'cp -rf ' + os.path.join(regression_path, 'image/') + ' ./ch_PP-OCRv3', 'mv ch_PP-OCRv3_det_infer.onnx ./ch_PP-OCRv3', 'mv ch_PP-OCRv3_rec_infer.onnx ./ch_PP-OCRv3', 'mv ch_PP-OCRv3_cls_infer.onnx ./ch_PP-OCRv3', 'mkdir ./ch_PP-OCRv3/workspace'] for str in mv_str_list: print(str) run(str, shell=True) def onnx2mlir(): transform_str_det = 'model_transform.py \ --model_name ch_PP-OCRv3_det \ --model_def ../ch_PP-OCRv3_det_infer.onnx \ --input_shapes [[1,3,960,608]] \ --mean 0.0,0.0,0.0 \ --scale 0.0039216,0.0039216,0.0039216 \ --keep_aspect_ratio \ --pixel_format rgb \ --output_names sigmoid_0.tmp_0 \ --test_input ../image/dog.jpg \ --test_result ch_PP-OCRv3_det_top_outputs.npz \ --mlir ./ch_PP-OCRv3_det.mlir' transform_str_rec = 'model_transform.py \ --model_name ch_PP-OCRv3_rec \ --model_def ../ch_PP-OCRv3_rec_infer.onnx \ --input_shapes [[1,3,48,320]] \ --mean 0.0,0.0,0.0 \ --scale 0.0039216,0.0039216,0.0039216 \ --keep_aspect_ratio \ --pixel_format rgb \ --output_names softmax_5.tmp_0 \ --test_input ../image/dog.jpg \ --test_result ch_PP-OCRv3_rec_top_outputs.npz \ --mlir ./ch_PP-OCRv3_rec.mlir' transform_str_cls = 'model_transform.py \ --model_name ch_PP-OCRv3_cls \ --model_def ../ch_PP-OCRv3_cls_infer.onnx \ --input_shapes [[1,3,48,192]] \ --mean 0.0,0.0,0.0 \ --scale 0.0039216,0.0039216,0.0039216 \ --keep_aspect_ratio \ --pixel_format rgb \ --output_names softmax_0.tmp_0 \ --test_input ../image/dog.jpg \ --test_result ch_PP-OCRv3_cls_top_outputs.npz \ --mlir ./ch_PP-OCRv3_cls.mlir' os.chdir('./ch_PP-OCRv3/workspace/') print(transform_str_det) run(transform_str_det, shell=True) print(transform_str_rec) run(transform_str_rec, shell=True) print(transform_str_cls) run(transform_str_cls, shell=True) os.chdir('../../') def mlir2bmodel(): det_str = 'model_deploy.py \ --mlir ./ch_PP-OCRv3_det.mlir \ --quantize F32 \ --chip bm1684x \ --test_input ./ch_PP-OCRv3_det_in_f32.npz \ --test_reference ./ch_PP-OCRv3_det_top_outputs.npz \ --model ./ch_PP-OCRv3_det_1684x_f32.bmodel' rec_str = 'model_deploy.py \ --mlir ./ch_PP-OCRv3_rec.mlir \ --quantize F32 \ --chip bm1684x \ --test_input ./ch_PP-OCRv3_rec_in_f32.npz \ --test_reference ./ch_PP-OCRv3_rec_top_outputs.npz \ --model ./ch_PP-OCRv3_rec_1684x_f32.bmodel' cls_str = 'model_deploy.py \ --mlir ./ch_PP-OCRv3_cls.mlir \ --quantize F32 \ --chip bm1684x \ --test_input ./ch_PP-OCRv3_cls_in_f32.npz \ --test_reference ./ch_PP-OCRv3_cls_top_outputs.npz \ --model ./ch_PP-OCRv3_cls_1684x_f32.bmodel' os.chdir('./ch_PP-OCRv3/workspace/') print(det_str) run(det_str, shell=True) print(rec_str) run(rec_str, shell=True) print(cls_str) run(cls_str, shell=True) os.chdir('../../') args = parse_arguments() if (args.auto): getPPOCRv3() fix_input_shape() paddle2onnx() mlir_prepare() onnx2mlir() mlir2bmodel() # 配置runtime，加载模型 runtime_option = fd.RuntimeOption() runtime_option.use_sophgo() # Detection模型, 检测文字框 det_model_file = './ch_PP-OCRv3/workspace/ch_PP-OCRv3_det_1684x_f32.bmodel' if args.auto else args.det_model det_params_file = "" # Classification模型，方向分类，可选 cls_model_file = './ch_PP-OCRv3/workspace/ch_PP-OCRv3_cls_1684x_f32.bmodel' if args.auto else args.cls_model cls_params_file = "" # Recognition模型，文字识别模型 rec_model_file = './ch_PP-OCRv3/workspace/ch_PP-OCRv3_rec_1684x_f32.bmodel' if args.auto else args.rec_model rec_params_file = "" rec_label_file = './ppocr_keys_v1.txt' if args.auto else args.rec_label_file image_file = './12.jpg' if args.auto else args.image # PPOCR的cls和rec模型现在已经支持推理一个Batch的数据 # 定义下面两个变量后, 可用于设置trt输入shape, 并在PPOCR模型初始化后, 完成Batch推理设置 cls_batch_size = 1 rec_batch_size = 1 # 当使用TRT时，分别给三个模型的runtime设置动态shape,并完成模型的创建. # 注意: 需要在检测模型创建完成后，再设置分类模型的动态输入并创建分类模型, 识别模型同理. # 如果用户想要自己改动检测模型的输入shape, 我们建议用户把检测模型的长和高设置为32的倍数. det_option = runtime_option det_option.set_trt_input_shape("x", [1, 3, 64, 64], [1, 3, 640, 640], [1, 3, 960, 960]) # 用户可以把TRT引擎文件保存至本地 # det_option.set_trt_cache_file(args.det_model + "/det_trt_cache.trt") det_model = fd.vision.ocr.DBDetector( det_model_file, det_params_file, runtime_option=det_option, model_format=fd.ModelFormat.SOPHGO) cls_option = runtime_option cls_option.set_trt_input_shape("x", [1, 3, 48, 10], [cls_batch_size, 3, 48, 320], [cls_batch_size, 3, 48, 1024]) # 用户可以把TRT引擎文件保存至本地 # cls_option.set_trt_cache_file(args.cls_model + "/cls_trt_cache.trt") cls_model = fd.vision.ocr.Classifier( cls_model_file, cls_params_file, runtime_option=cls_option, model_format=fd.ModelFormat.SOPHGO) rec_option = runtime_option rec_option.set_trt_input_shape("x", [1, 3, 48, 10], [rec_batch_size, 3, 48, 320], [rec_batch_size, 3, 48, 2304]) # 用户可以把TRT引擎文件保存至本地 # rec_option.set_trt_cache_file(args.rec_model + "/rec_trt_cache.trt") rec_model = fd.vision.ocr.Recognizer( rec_model_file, rec_params_file, rec_label_file, runtime_option=rec_option, model_format=fd.ModelFormat.SOPHGO) # 创建PP-OCR，串联3个模型，其中cls_model可选，如无需求，可设置为None ppocr_v3 = fd.vision.ocr.PPOCRv3( det_model=det_model, cls_model=cls_model, rec_model=rec_model) # 需要使用下行代码, 来启用rec模型的静态shape推理，这里rec模型的静态输入为[3, 48, 320] rec_model.preprocessor.static_shape_infer = True rec_model.preprocessor.rec_image_shape = [3, 48, 320] # 给cls和rec模型设置推理时的batch size # 此值能为-1, 和1到正无穷 # 当此值为-1时, cls和rec模型的batch size将默认和det模型检测出的框的数量相同 ppocr_v3.cls_batch_size = cls_batch_size ppocr_v3.rec_batch_size = rec_batch_size # 预测图片准备 im = cv2.imread(image_file) #预测并打印结果 result = ppocr_v3.predict(im) print(result) # 可视化结果 vis_im = fd.vision.vis_ppocr(im, result) cv2.imwrite("sophgo_result.jpg", vis_im) print("Visualized result save in ./sophgo_result.jpg")