From cf00f769deb13173a9ce2d07ed4be9ffd75247e4 Mon Sep 17 00:00:00 2001 From: SWHL Date: Sun, 25 Sep 2022 16:41:31 +0800 Subject: [PATCH] Update python/ort-gpu docs --- ocrweb/api.py | 1 - ocrweb/config.yaml | 1 + .../ch_ppocr_v2_cls/utils.py | 9 ++- .../ch_ppocr_v3_det/utils.py | 57 ++++++++++++++----- .../ch_ppocr_v3_rec/utils.py | 12 ++-- python/README.md | 33 ++++------- .../ch_ppocr_v2_cls/utils.py | 4 +- .../ch_ppocr_v3_det/utils.py | 4 +- .../ch_ppocr_v3_rec/utils.py | 4 +- 9 files changed, 79 insertions(+), 46 deletions(-) diff --git a/ocrweb/api.py b/ocrweb/api.py index bdbea45..e9a9009 100644 --- a/ocrweb/api.py +++ b/ocrweb/api.py @@ -3,7 +3,6 @@ # @Contact: liekkaskono@163.com import base64 from wsgiref.simple_server import make_server -import json import cv2 import numpy as np diff --git a/ocrweb/config.yaml b/ocrweb/config.yaml index 9f40a0d..5b84273 100644 --- a/ocrweb/config.yaml +++ b/ocrweb/config.yaml @@ -37,6 +37,7 @@ Det: max_candidates: 1000 unclip_ratio: 1.6 use_dilation: true + score_mode: fast Cls: module_name: ch_ppocr_v2_cls diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py index 8ee0fa5..93f09ba 100644 --- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py +++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py @@ -15,7 +15,8 @@ import warnings import yaml from onnxruntime import (get_available_providers, get_device, - SessionOptions, InferenceSession) + SessionOptions, InferenceSession, + GraphOptimizationLevel) class OrtInferSession(object): @@ -23,15 +24,19 @@ class OrtInferSession(object): sess_opt = SessionOptions() sess_opt.log_severity_level = 4 sess_opt.enable_cpu_mem_arena = False + sess_opt.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL cuda_ep = 'CUDAExecutionProvider' cpu_ep = 'CPUExecutionProvider' + cpu_provider_options = { + "arena_extend_strategy": "kSameAsRequested", + } EP_list = [] if config['use_cuda'] and get_device() == 'GPU' \ and cuda_ep in get_available_providers(): EP_list = [(cuda_ep, config[cuda_ep])] - EP_list.append(cpu_ep) + EP_list.append((cpu_ep, cpu_provider_options)) self.session = InferenceSession(config['model_path'], sess_options=sess_opt, diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py index 674ed8b..75e7bb7 100644 --- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py +++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py @@ -26,7 +26,8 @@ import six import yaml from shapely.geometry import Polygon from onnxruntime import (get_available_providers, get_device, - SessionOptions, InferenceSession) + SessionOptions, InferenceSession, + GraphOptimizationLevel) class OrtInferSession(object): @@ -34,15 +35,19 @@ class OrtInferSession(object): sess_opt = SessionOptions() sess_opt.log_severity_level = 4 sess_opt.enable_cpu_mem_arena = False + sess_opt.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL cuda_ep = 'CUDAExecutionProvider' cpu_ep = 'CPUExecutionProvider' + cpu_provider_options = { + "arena_extend_strategy": "kSameAsRequested", + } EP_list = [] if config['use_cuda'] and get_device() == 'GPU' \ and cuda_ep in get_available_providers(): EP_list = [(cuda_ep, config[cuda_ep])] - EP_list.append(cpu_ep) + EP_list.append((cpu_ep, cpu_provider_options)) self.session = InferenceSession(config['model_path'], sess_options=sess_opt, @@ -152,19 +157,19 @@ class DetResizeForTest(object): self.image_shape = kwargs['image_shape'] self.resize_type = 1 elif 'limit_side_len' in kwargs: - self.limit_side_len = kwargs['limit_side_len'] + self.limit_side_len = kwargs.get('limit_side_len', 736) self.limit_type = kwargs.get('limit_type', 'min') if 'resize_long' in kwargs: self.resize_type = 2 self.resize_long = kwargs.get('resize_long', 960) else: - self.limit_side_len = 736 - self.limit_type = 'min' + self.limit_side_len = kwargs.get('limit_side_len', 736) + self.limit_type = kwargs.get('limit_type', 'min') def __call__(self, data): img = data['image'] - src_h, src_w, _ = img.shape + src_h, src_w = img.shape[:2] if self.resize_type == 0: # img, shape = self.resize_image_type0(img) @@ -196,7 +201,7 @@ class DetResizeForTest(object): img, (ratio_h, ratio_w) """ limit_side_len = self.limit_side_len - h, w, _ = img.shape + h, w = img.shape[:2] # limit the max side if self.limit_type == 'max': @@ -233,7 +238,7 @@ class DetResizeForTest(object): return img, [ratio_h, ratio_w] def resize_image_type2(self, img): - h, w, _ = img.shape + h, w = img.shape[:2] resize_w = w resize_h = h @@ -299,12 +304,14 @@ class DBPostProcess(object): box_thresh=0.7, max_candidates=1000, unclip_ratio=2.0, + score_mode="fast", use_dilation=False): self.thresh = thresh self.box_thresh = box_thresh self.max_candidates = max_candidates self.unclip_ratio = unclip_ratio self.min_size = 3 + self.score_mode = score_mode if use_dilation: self.dilation_kernel = np.array([[1, 1], [1, 1]]) @@ -337,7 +344,10 @@ class DBPostProcess(object): if sside < self.min_size: continue points = np.array(points) - score = self.box_score_fast(pred, points.reshape(-1, 2)) + if self.score_mode == "fast": + score = self.box_score_fast(pred, points.reshape(-1, 2)) + else: + score = self.box_score_slow(pred, contour) if self.box_thresh > score: continue @@ -390,10 +400,10 @@ class DBPostProcess(object): def box_score_fast(self, bitmap, _box): h, w = bitmap.shape[:2] box = _box.copy() - xmin = np.clip(np.floor(box[:, 0].min()).astype(np.int), 0, w - 1) - xmax = np.clip(np.ceil(box[:, 0].max()).astype(np.int), 0, w - 1) - ymin = np.clip(np.floor(box[:, 1].min()).astype(np.int), 0, h - 1) - ymax = np.clip(np.ceil(box[:, 1].max()).astype(np.int), 0, h - 1) + xmin = np.clip(np.floor(box[:, 0].min()).astype(np.int32), 0, w - 1) + xmax = np.clip(np.ceil(box[:, 0].max()).astype(np.int32), 0, w - 1) + ymin = np.clip(np.floor(box[:, 1].min()).astype(np.int32), 0, h - 1) + ymax = np.clip(np.ceil(box[:, 1].max()).astype(np.int32), 0, h - 1) mask = np.zeros((ymax - ymin + 1, xmax - xmin + 1), dtype=np.uint8) box[:, 0] = box[:, 0] - xmin @@ -401,6 +411,27 @@ class DBPostProcess(object): cv2.fillPoly(mask, box.reshape(1, -1, 2).astype(np.int32), 1) return cv2.mean(bitmap[ymin:ymax + 1, xmin:xmax + 1], mask)[0] + def box_score_slow(self, bitmap, contour): + ''' + box_score_slow: use polyon mean score as the mean score + ''' + h, w = bitmap.shape[:2] + contour = contour.copy() + contour = np.reshape(contour, (-1, 2)) + + xmin = np.clip(np.min(contour[:, 0]), 0, w - 1) + xmax = np.clip(np.max(contour[:, 0]), 0, w - 1) + ymin = np.clip(np.min(contour[:, 1]), 0, h - 1) + ymax = np.clip(np.max(contour[:, 1]), 0, h - 1) + + mask = np.zeros((ymax - ymin + 1, xmax - xmin + 1), dtype=np.uint8) + + contour[:, 0] = contour[:, 0] - xmin + contour[:, 1] = contour[:, 1] - ymin + + cv2.fillPoly(mask, contour.reshape(1, -1, 2).astype(np.int32), 1) + return cv2.mean(bitmap[ymin:ymax + 1, xmin:xmax + 1], mask)[0] + def __call__(self, pred, shape_list): pred = pred[:, 0, :, :] segmentation = pred > self.thresh diff --git a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py index 8f697c1..7576fcf 100644 --- a/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py +++ b/ocrweb/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py @@ -6,7 +6,8 @@ import warnings import numpy as np import yaml from onnxruntime import (get_available_providers, get_device, - SessionOptions, InferenceSession) + SessionOptions, InferenceSession, + GraphOptimizationLevel) class OrtInferSession(object): @@ -14,16 +15,19 @@ class OrtInferSession(object): sess_opt = SessionOptions() sess_opt.log_severity_level = 4 sess_opt.enable_cpu_mem_arena = False + sess_opt.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL cuda_ep = 'CUDAExecutionProvider' cpu_ep = 'CPUExecutionProvider' + cpu_provider_options = { + "arena_extend_strategy": "kSameAsRequested", + } EP_list = [] if config['use_cuda'] and get_device() == 'GPU' \ and cuda_ep in get_available_providers(): EP_list = [(cuda_ep, config[cuda_ep])] - - EP_list.append(cpu_ep) + EP_list.append((cpu_ep, cpu_provider_options)) self.session = InferenceSession(config['model_path'], sess_options=sess_opt, @@ -120,5 +124,5 @@ class CTCLabelDecode(object): else: conf_list.append(1) text = ''.join(char_list) - result_list.append((text, np.mean(conf_list + [1e-20]))) + result_list.append((text, np.mean(conf_list + [1e-10] ))) return result_list diff --git a/python/README.md b/python/README.md index ce4e24f..5caaf64 100755 --- a/python/README.md +++ b/python/README.md @@ -277,29 +277,16 @@ print(rec_res) python tests/benchmark/benchmark.py --yaml_path config_gpu.yaml ``` 3. 运行相关信息汇总：（以下仅为个人测试情况，具体情况请自行测试） - - - 来自[zhsunlight](https://github.com/zhsunlight)的测试，感谢 - - 设备型号：宏碁(Acer) 暗影骑士·威N50-N93游戏台式机 - - OS: Windows - - CPU型号：十代i5-10400F 16G 512G SSD - - GPU型号：NVIDIA GeForce GTX 1660Super 6G - - onnxruntime-gpu: 1.11.0 - - 耗时情况： - |设备|总耗时(s)|平均耗时(s/img)| - |:---:|:---:|:---:| - |CPU|296.8841|1.18282| - |GPU|646.14667|2.57429| - - 来自[SWHL](https://github.com/SWHL)的测试 - - 设备型号：Docker - - OS: Linux Ubuntu - - CPU型号：- - - GPU型号：NVIDIA V100S 16G - - onnxruntime-gpu: 1.7.0 - - 耗时情况： - |设备|总耗时(s)|平均耗时(s/img)| - |:---:|:---:|:---:| - |CPU|1079.4726|4.3001| - |GPU|525.8244|2.0989| + - 环境 + |测试者|设备|OS|CPU|GPU|onnxruntime-gpu| + |:--|:--|:--|:--|:--|:--| + |[1][zhsunlight](https://github.com/zhsunlight)|宏碁(Acer) 暗影骑士·威N50-N93游戏台式机|Windows|十代i5-10400F 16G 512G SSD|NVIDIA GeForce GTX 1660Super 6G|1.11.0| + |[2][SWHL](https://github.com/SWHL)|服务器|Linux|-|NVIDIA GeForce RTX 3090|1.12.1| + - 耗时 + |对应上面序号|CPU总耗时(s)|CPU平均耗时(s/img)|GPU总耗时(s)|GPU平均耗时(s/img)|| + |:---:|:---:|:---:|:---:|:---:|:---:| + |[1]|296.8841|1.18282|646.14667|2.57429| + |[2]|149.35427|0.50504|250.81760|0.99927| ### OpenVINO GPU推理配置 - 官方参考文档：[docs](https://docs.openvino.ai/latest/api/ie_python_api/_autosummary/openvino.runtime.Core.html?highlight=compile_model#openvino.runtime.Core.compile_model) diff --git a/python/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py b/python/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py index d44c08b..93f09ba 100644 --- a/python/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py +++ b/python/rapidocr_onnxruntime/ch_ppocr_v2_cls/utils.py @@ -15,7 +15,8 @@ import warnings import yaml from onnxruntime import (get_available_providers, get_device, - SessionOptions, InferenceSession) + SessionOptions, InferenceSession, + GraphOptimizationLevel) class OrtInferSession(object): @@ -23,6 +24,7 @@ class OrtInferSession(object): sess_opt = SessionOptions() sess_opt.log_severity_level = 4 sess_opt.enable_cpu_mem_arena = False + sess_opt.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL cuda_ep = 'CUDAExecutionProvider' cpu_ep = 'CPUExecutionProvider' diff --git a/python/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py b/python/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py index c108a97..75e7bb7 100644 --- a/python/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py +++ b/python/rapidocr_onnxruntime/ch_ppocr_v3_det/utils.py @@ -26,7 +26,8 @@ import six import yaml from shapely.geometry import Polygon from onnxruntime import (get_available_providers, get_device, - SessionOptions, InferenceSession) + SessionOptions, InferenceSession, + GraphOptimizationLevel) class OrtInferSession(object): @@ -34,6 +35,7 @@ class OrtInferSession(object): sess_opt = SessionOptions() sess_opt.log_severity_level = 4 sess_opt.enable_cpu_mem_arena = False + sess_opt.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL cuda_ep = 'CUDAExecutionProvider' cpu_ep = 'CPUExecutionProvider' diff --git a/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py b/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py index 392e659..7576fcf 100644 --- a/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py +++ b/python/rapidocr_onnxruntime/ch_ppocr_v3_rec/utils.py @@ -6,7 +6,8 @@ import warnings import numpy as np import yaml from onnxruntime import (get_available_providers, get_device, - SessionOptions, InferenceSession) + SessionOptions, InferenceSession, + GraphOptimizationLevel) class OrtInferSession(object): @@ -14,6 +15,7 @@ class OrtInferSession(object): sess_opt = SessionOptions() sess_opt.log_severity_level = 4 sess_opt.enable_cpu_mem_arena = False + sess_opt.graph_optimization_level = GraphOptimizationLevel.ORT_ENABLE_ALL cuda_ep = 'CUDAExecutionProvider' cpu_ep = 'CPUExecutionProvider'