From 9059f747e62bb0ffe2a3dc06f011627eed56c882 Mon Sep 17 00:00:00 2001 From: andyj <87074272+andyjpaddle@users.noreply.github.com> Date: Mon, 30 Jan 2023 16:05:14 +0800 Subject: [PATCH] support new version of numpy (#8921) * add finetune en doc & test=document_fix * fix dead link & test=document_fix * fix dead link & test=document_fix * update check img * fix det res dtype * update args default type & test=document_fix * fix numpy version * support numpy1.24.0 --- ...345\215\241\350\257\201\347\261\273OCR.md" | 24 +++++++++---------- deploy/avh/convert_image.py | 15 ++++++------ deploy/hubserving/ocr_det/module.py | 2 +- deploy/hubserving/ocr_system/module.py | 2 +- ppocr/data/imaug/abinet_aug.py | 13 +++++----- ppocr/data/imaug/fce_aug.py | 2 +- ppocr/data/imaug/fce_targets.py | 17 +++++++------ ppocr/data/imaug/label_ops.py | 6 ++--- ppocr/modeling/heads/proposal_local_graph.py | 2 +- ppocr/postprocess/drrg_postprocess.py | 7 +++--- ppocr/postprocess/fce_postprocess.py | 4 ++-- 11 files changed, 48 insertions(+), 46 deletions(-) diff --git "a/applications/\345\277\253\351\200\237\346\236\204\345\273\272\345\215\241\350\257\201\347\261\273OCR.md" "b/applications/\345\277\253\351\200\237\346\236\204\345\273\272\345\215\241\350\257\201\347\261\273OCR.md" index a5a9460f..79266c6c 100644 --- "a/applications/\345\277\253\351\200\237\346\236\204\345\273\272\345\215\241\350\257\201\347\261\273OCR.md" +++ "b/applications/\345\277\253\351\200\237\346\236\204\345\273\272\345\215\241\350\257\201\347\261\273OCR.md" @@ -223,7 +223,7 @@ AIStudio项目链接：[快速构建卡证类OCR](https://aistudio.baidu.com/ais 2）获取并解压预训练模型，如果要使用其他模型可以从模型库里自主选择合适模型。 ``` -!wget -P work/pre_trained/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar +!wget -P work/pre_trained/ https://paddleocr.bj.bcebos.com/PP-OCRv3/chinese/ch_PP-OCRv3_det_distill_train.tar !tar -vxf /home/aistudio/work/pre_trained/ch_PP-OCRv3_det_distill_train.tar -C /home/aistudio/work/pre_trained ``` 3） 安装必要依赖 @@ -275,7 +275,7 @@ AIStudio项目链接：[快速构建卡证类OCR](https://aistudio.baidu.com/ais ```python class DetLabelEncode(object): - + # 修改检测标签的编码处，新增了参数分类数：num_classes，重写初始化方法，以及分类标签的读取 def __init__(self, label_list, num_classes=8, **kwargs): @@ -315,11 +315,11 @@ class DetLabelEncode(object): classes.append(int(self.label_list.index(txt))) if len(boxes) == 0: - + return None boxes = self.expand_points_num(boxes) boxes = np.array(boxes, dtype=np.float32) - txt_tags = np.array(txt_tags, dtype=np.bool) + txt_tags = np.array(txt_tags, dtype=np.bool_) classes = classes data['polys'] = boxes data['texts'] = txts @@ -410,10 +410,10 @@ class MakeShrinkMap(object): data['shrink_map'] = gt - + if self.num_classes > 1: data['class_mask'] = gt_class - + data['shrink_mask'] = mask return data ``` @@ -634,10 +634,10 @@ class DBPostProcess(object): ''' 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 @@ -752,11 +752,11 @@ class DBPostProcess(object): 其他命令： ``` !python /home/aistudio/work/PaddleOCR/tools/eval.py -c /home/aistudio/work/PaddleOCR/configs/det/det_mv3_db.yml -!python /home/aistudio/work/PaddleOCR/tools/infer_det.py -c /home/aistudio/work/PaddleOCR/configs/det/det_mv3_db.yml +!python /home/aistudio/work/PaddleOCR/tools/infer_det.py -c /home/aistudio/work/PaddleOCR/configs/det/det_mv3_db.yml ``` 模型推理 ``` -!python /home/aistudio/work/PaddleOCR/tools/infer/predict_det.py --image_dir="/home/aistudio/work/test_img/" --det_model_dir="/home/aistudio/work/PaddleOCR/output/infer" +!python /home/aistudio/work/PaddleOCR/tools/infer/predict_det.py --image_dir="/home/aistudio/work/test_img/" --det_model_dir="/home/aistudio/work/PaddleOCR/output/infer" ``` ## 5 总结 diff --git a/deploy/avh/convert_image.py b/deploy/avh/convert_image.py index 747ab29e..7c6dbd7f 100755 --- a/deploy/avh/convert_image.py +++ b/deploy/avh/convert_image.py @@ -24,6 +24,7 @@ import math from PIL import Image import numpy as np + def resize_norm_img(img, image_shape, padding=True): imgC, imgH, imgW = image_shape h = img.shape[0] @@ -61,9 +62,8 @@ def create_header_file(name, tensor_name, tensor_data, output_path): raw_path = file_path.with_suffix(".h").resolve() with open(raw_path, "w") as header_file: header_file.write( - "\n" - + f"const size_t {tensor_name}_len = {tensor_data.size};\n" - + f'__attribute__((section(".data.tvm"), aligned(16))) float {tensor_name}[] = ' + "\n" + f"const size_t {tensor_name}_len = {tensor_data.size};\n" + + f'__attribute__((section(".data.tvm"), aligned(16))) float {tensor_name}[] = ' ) header_file.write("{") @@ -80,22 +80,21 @@ def create_headers(image_name): # Resize image to 32x320 img = cv2.imread(img_path) - img = resize_norm_img(img, [3,32,320]) + img = resize_norm_img(img, [3, 32, 320]) img_data = img.astype("float32") - + # # Add the batch dimension, as we are expecting 4-dimensional input: NCHW. img_data = np.expand_dims(img_data, axis=0) # Create input header file create_header_file("inputs", "input", img_data, "./include") # Create output header file - output_data = np.zeros([7760], np.float) + output_data = np.zeros([7760], np.float32) create_header_file( "outputs", "output", output_data, - "./include", - ) + "./include", ) if __name__ == "__main__": diff --git a/deploy/hubserving/ocr_det/module.py b/deploy/hubserving/ocr_det/module.py index 8fef3be0..3dbaf161 100644 --- a/deploy/hubserving/ocr_det/module.py +++ b/deploy/hubserving/ocr_det/module.py @@ -122,7 +122,7 @@ class OCRDet(hub.Module): rec_res_final = [] for dno in range(len(dt_boxes)): rec_res_final.append({ - 'text_region': dt_boxes[dno].astype(np.int).tolist() + 'text_region': dt_boxes[dno].astype(np.int32).tolist() }) all_results.append(rec_res_final) return all_results diff --git a/deploy/hubserving/ocr_system/module.py b/deploy/hubserving/ocr_system/module.py index dff3abb4..192fff96 100644 --- a/deploy/hubserving/ocr_system/module.py +++ b/deploy/hubserving/ocr_system/module.py @@ -130,7 +130,7 @@ class OCRSystem(hub.Module): rec_res_final.append({ 'text': text, 'confidence': float(score), - 'text_region': dt_boxes[dno].astype(np.int).tolist() + 'text_region': dt_boxes[dno].astype(np.int32).tolist() }) all_results.append(rec_res_final) return all_results diff --git a/ppocr/data/imaug/abinet_aug.py b/ppocr/data/imaug/abinet_aug.py index 1b93751b..bcbdadb1 100644 --- a/ppocr/data/imaug/abinet_aug.py +++ b/ppocr/data/imaug/abinet_aug.py @@ -205,7 +205,7 @@ class CVRandomAffine(object): for x, y in startpoints] rect = cv2.minAreaRect(np.array(endpoints)) - bbox = cv2.boxPoints(rect).astype(dtype=np.int) + bbox = cv2.boxPoints(rect).astype(dtype=np.int32) max_x, max_y = bbox[:, 0].max(), bbox[:, 1].max() min_x, min_y = bbox[:, 0].min(), bbox[:, 1].min() @@ -234,9 +234,9 @@ class CVRandomPerspective(object): def get_params(self, width, height, distortion): offset_h = sample_asym( - distortion * height / 2, size=4).astype(dtype=np.int) + distortion * height / 2, size=4).astype(dtype=np.int32) offset_w = sample_asym( - distortion * width / 2, size=4).astype(dtype=np.int) + distortion * width / 2, size=4).astype(dtype=np.int32) topleft = (offset_w[0], offset_h[0]) topright = (width - 1 - offset_w[1], offset_h[1]) botright = (width - 1 - offset_w[2], height - 1 - offset_h[2]) @@ -256,7 +256,7 @@ class CVRandomPerspective(object): # TODO: more robust way to crop image rect = cv2.minAreaRect(endpoints) - bbox = cv2.boxPoints(rect).astype(dtype=np.int) + bbox = cv2.boxPoints(rect).astype(dtype=np.int32) max_x, max_y = bbox[:, 0].max(), bbox[:, 1].max() min_x, min_y = bbox[:, 0].min(), bbox[:, 1].min() min_x, min_y = max(min_x, 0), max(min_y, 0) @@ -441,7 +441,8 @@ class SVTRGeometry(object): self.p = p self.transforms = [] self.transforms.append(CVRandomRotation(degrees=degrees)) - self.transforms.append(CVRandomAffine( + self.transforms.append( + CVRandomAffine( degrees=degrees, translate=translate, scale=scale, shear=shear)) self.transforms.append(CVRandomPerspective(distortion=distortion)) @@ -455,4 +456,4 @@ class SVTRGeometry(object): img = self.transforms[random.randint(0, 2)](img) return img else: - return img \ No newline at end of file + return img diff --git a/ppocr/data/imaug/fce_aug.py b/ppocr/data/imaug/fce_aug.py index 66bafef1..baaaa335 100644 --- a/ppocr/data/imaug/fce_aug.py +++ b/ppocr/data/imaug/fce_aug.py @@ -208,7 +208,7 @@ class RandomCropFlip: for polygon in all_polys: rect = cv2.minAreaRect(polygon.astype(np.int32).reshape(-1, 2)) box = cv2.boxPoints(rect) - box = np.int0(box) + box = np.int64(box) text_polys.append([box[0], box[1], box[2], box[3]]) polys = np.array(text_polys, dtype=np.int32) diff --git a/ppocr/data/imaug/fce_targets.py b/ppocr/data/imaug/fce_targets.py index 8c64276e..054631cb 100644 --- a/ppocr/data/imaug/fce_targets.py +++ b/ppocr/data/imaug/fce_targets.py @@ -22,10 +22,12 @@ from numpy.fft import fft from numpy.linalg import norm import sys + def vector_slope(vec): assert len(vec) == 2 return abs(vec[1] / (vec[0] + 1e-8)) + class FCENetTargets: """Generate the ground truth targets of FCENet: Fourier Contour Embedding for Arbitrary-Shaped Text Detection. @@ -107,7 +109,9 @@ class FCENetTargets: for i in range(1, n): current_line_len = i * delta_length - while current_edge_ind + 1 < len(length_cumsum) and current_line_len >= length_cumsum[current_edge_ind + 1]: + while current_edge_ind + 1 < len( + length_cumsum) and current_line_len >= length_cumsum[ + current_edge_ind + 1]: current_edge_ind += 1 current_edge_end_shift = current_line_len - length_cumsum[ @@ -239,10 +243,9 @@ class FCENetTargets: head_inds = [head_start, head_end] tail_inds = [tail_start, tail_end] else: - if vector_slope(points[1] - points[0]) + vector_slope( - points[3] - points[2]) < vector_slope(points[ - 2] - points[1]) + vector_slope(points[0] - points[ - 3]): + if vector_slope(points[1] - points[0]) + vector_slope(points[ + 3] - points[2]) < vector_slope(points[2] - points[ + 1]) + vector_slope(points[0] - points[3]): horizontal_edge_inds = [[0, 1], [2, 3]] vertical_edge_inds = [[3, 0], [1, 2]] else: @@ -582,7 +585,7 @@ class FCENetTargets: lv_ignore_polys = [[] for i in range(len(lv_size_divs))] level_maps = [] for poly in text_polys: - polygon = np.array(poly, dtype=np.int).reshape((1, -1, 2)) + polygon = np.array(poly, dtype=np.int32).reshape((1, -1, 2)) _, _, box_w, box_h = cv2.boundingRect(polygon) proportion = max(box_h, box_w) / (h + 1e-8) @@ -591,7 +594,7 @@ class FCENetTargets: lv_text_polys[ind].append(poly / lv_size_divs[ind]) for ignore_poly in ignore_polys: - polygon = np.array(ignore_poly, dtype=np.int).reshape((1, -1, 2)) + polygon = np.array(ignore_poly, dtype=np.int32).reshape((1, -1, 2)) _, _, box_w, box_h = cv2.boundingRect(polygon) proportion = max(box_h, box_w) / (h + 1e-8) diff --git a/ppocr/data/imaug/label_ops.py b/ppocr/data/imaug/label_ops.py index 8f56ea10..b64dc076 100644 --- a/ppocr/data/imaug/label_ops.py +++ b/ppocr/data/imaug/label_ops.py @@ -64,7 +64,7 @@ class DetLabelEncode(object): return None boxes = self.expand_points_num(boxes) boxes = np.array(boxes, dtype=np.float32) - txt_tags = np.array(txt_tags, dtype=np.bool) + txt_tags = np.array(txt_tags, dtype=np.bool_) data['polys'] = boxes data['texts'] = txts @@ -218,7 +218,7 @@ class E2ELabelEncodeTest(BaseRecLabelEncode): else: txt_tags.append(False) boxes = np.array(boxes, dtype=np.float32) - txt_tags = np.array(txt_tags, dtype=np.bool) + txt_tags = np.array(txt_tags, dtype=np.bool_) data['polys'] = boxes data['ignore_tags'] = txt_tags temp_texts = [] @@ -254,7 +254,7 @@ class E2ELabelEncodeTrain(object): else: txt_tags.append(False) boxes = np.array(boxes, dtype=np.float32) - txt_tags = np.array(txt_tags, dtype=np.bool) + txt_tags = np.array(txt_tags, dtype=np.bool_) data['polys'] = boxes data['texts'] = txts diff --git a/ppocr/modeling/heads/proposal_local_graph.py b/ppocr/modeling/heads/proposal_local_graph.py index 7887c4ff..a4865613 100644 --- a/ppocr/modeling/heads/proposal_local_graph.py +++ b/ppocr/modeling/heads/proposal_local_graph.py @@ -40,7 +40,7 @@ def fill_hole(input_mask): mask = np.zeros((h + 4, w + 4), np.uint8) cv2.floodFill(canvas, mask, (0, 0), 1) - canvas = canvas[1:h + 1, 1:w + 1].astype(np.bool) + canvas = canvas[1:h + 1, 1:w + 1].astype(np.bool_) return ~canvas | input_mask diff --git a/ppocr/postprocess/drrg_postprocess.py b/ppocr/postprocess/drrg_postprocess.py index 353081c9..56fd034f 100644 --- a/ppocr/postprocess/drrg_postprocess.py +++ b/ppocr/postprocess/drrg_postprocess.py @@ -68,7 +68,7 @@ def graph_propagation(edges, scores, text_comps, edge_len_thr=50.): score_dict[edge[0], edge[1]] = scores[i] nodes = np.sort(np.unique(edges.flatten())) - mapping = -1 * np.ones((np.max(nodes) + 1), dtype=np.int) + mapping = -1 * np.ones((np.max(nodes) + 1), dtype=np.int32) mapping[nodes] = np.arange(nodes.shape[0]) order_inds = mapping[edges] vertices = [Node(node) for node in nodes] @@ -93,9 +93,8 @@ def connected_components(nodes, score_dict, link_thr): while node_queue: node = node_queue.pop(0) neighbors = set([ - neighbor for neighbor in node.links - if score_dict[tuple(sorted([node.ind, neighbor.ind]))] >= - link_thr + neighbor for neighbor in node.links if + score_dict[tuple(sorted([node.ind, neighbor.ind]))] >= link_thr ]) neighbors.difference_update(cluster) nodes.difference_update(neighbors) diff --git a/ppocr/postprocess/fce_postprocess.py b/ppocr/postprocess/fce_postprocess.py index 8e0716f9..959f86ef 100755 --- a/ppocr/postprocess/fce_postprocess.py +++ b/ppocr/postprocess/fce_postprocess.py @@ -31,7 +31,7 @@ def fill_hole(input_mask): mask = np.zeros((h + 4, w + 4), np.uint8) cv2.floodFill(canvas, mask, (0, 0), 1) - canvas = canvas[1:h + 1, 1:w + 1].astype(np.bool) + canvas = canvas[1:h + 1, 1:w + 1].astype(np.bool_) return ~canvas | input_mask @@ -234,7 +234,7 @@ class FCEPostProcess(object): poly = np.array(boundary[:-1]).reshape(-1, 2).astype(np.float32) score = boundary[-1] points = cv2.boxPoints(cv2.minAreaRect(poly)) - points = np.int0(points) + points = np.int64(points) new_boundaries.append(points.reshape(-1).tolist() + [score]) boundaries = new_boundaries -- GitLab