pybaseutils 2.0.2__tar.gz → 2.0.5__tar.gz

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: pybaseutils
-Version: 2.0.2
+Version: 2.0.5
 Summary: pybaseutils
 Home-page: https://github.com/PanJinquan/base-utils
 Author: PanJinquan

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/__init__.py

@@ -5,4 +5,4 @@
     @Date   : 2019-05-07 17:40:27
 """
 
-__version__ = '2.0.2'
+__version__ = '2.0.5'

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/converter/convert_cvat2labelme.py

@@ -38,14 +38,15 @@ def parser_annotations(xml_file):
     return filename, annos, width, height
 
 
-def convert_cvat2labelme(anno_dir, image_dir="", vis=False):
+def convert_cvat2labelme(anno_dir, image_dir="", out_dir="", thickness=1, fontScale=1.0, vis=False):
     """
     将CVAT标注格式(LabelMe 3.0)转换labelme通用格式
-    :param anno_dir: 标注文件目录，输出json文件与anno_dir同目录
+    :param anno_dir: 标注文件目录，默认输出json文件与anno_dir同目录
     :param image_dir: 图片文件目录
     :param vis:
     :return:
     """
+    if not out_dir: out_dir = os.path.join(os.path.dirname(anno_dir), "json")
     xml_list = file_utils.get_files_lists(anno_dir, postfix=["*.xml"])
     for xml_file in tqdm(xml_list):
         image_name, annos, w, h = parser_annotations(xml_file)
@@ -56,13 +57,15 @@ def convert_cvat2labelme(anno_dir, image_dir="", vis=False):
             continue
         if image_dir or vis:
             image_file = os.path.join(image_dir, image_name)
+            print(image_file, "labels:{}".format(labels))
             image = cv2.imread(image_file)
             h, w = image.shape[:2]
             if vis:
-                image = image_utils.draw_image_contours(image, points, texts=labels)
+                image = image_utils.draw_image_contours(image, points, texts=labels, thickness=thickness,
+                                                        fontScale=fontScale)
                 image_utils.cv_show_image("det", image)
         image_id = image_name.split(".")[0]
-        json_file = os.path.join(anno_dir, f"{image_id}.json")
+        json_file = os.path.join(out_dir, f"{image_id}.json")
         build_labelme.maker_labelme(json_file, points, labels, image_name, image_size=[w, h], image_bs64=None)
 
 

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/converter/convert_voc2yolo.py

@@ -80,7 +80,7 @@ def convert_voc2yolo_file(filename, out_text_dir, out_image_dir=None, class_name
             format = "jpg"
         newname = "{}.{}".format(image_id, format)
         text_path = file_utils.create_dir(out_text_dir, None, "{}.txt".format(image_id))
-        labels = np.asarray(labels).astype(np.int).reshape(-1)
+        labels = np.asarray(labels).astype(np.int32).reshape(-1)
         contents = create_yolo_dets(bboxes, labels, width=w, height=h)
         file_utils.write_data(text_path, contents)
         if out_image_dir:
@@ -131,7 +131,7 @@ def convert_voc2yolo(data_root, out_text_dir, out_image_dir=None, class_name=Non
             format = "jpg"
         newname = "{}.{}".format(image_id, format)
         text_path = file_utils.create_dir(out_text_dir, None, "{}.txt".format(image_id))
-        labels = np.asarray(labels).astype(np.int).reshape(-1)
+        labels = np.asarray(labels).astype(np.int32).reshape(-1)
         contents = create_yolo_dets(bboxes, labels, width=w, height=h)
         file_utils.write_data(text_path, contents, split=" ")
         if out_image_dir:
@@ -183,7 +183,7 @@ def convert_voc2yolo_segment(data_root, out_text_dir, out_image_dir=None, class_
             format = "jpg"
         newname = "{}.{}".format(image_id, format)
         text_path = file_utils.create_dir(out_text_dir, None, "{}.txt".format(image_id))
-        labels = np.asarray(labels).astype(np.int).reshape(-1)
+        labels = np.asarray(labels).astype(np.int32).reshape(-1)
         contents = create_yolo_segs(bboxes, labels, width=w, height=h)
         file_utils.write_data(text_path, contents, split=" ")
         if out_image_dir:

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/converter/convert_yolo2voc.py

@@ -50,7 +50,7 @@ def convert_yolo2voc(filename, out_xml_dir=None, out_image_dir=None, class_name=
             format = "jpg"
         newname = "{}.{}".format(image_id, format)
         xml_path = file_utils.create_dir(out_xml_dir, None, "{}.xml".format(image_id))
-        labels = np.asarray(labels).astype(np.int).reshape(-1)
+        labels = np.asarray(labels).astype(np.int32).reshape(-1)
         objects = build_voc.create_objects(bboxes, labels, keypoints=None, class_name=class_name)
         build_voc.write_voc_xml_objects(newname, image_shape, objects, xml_path)
         if out_image_dir:

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/coords_utils.py

@@ -19,6 +19,7 @@ def clip_xyxy(xyxy: np.ndarray, valid_range):
     :param valid_range:有效范围(xmin,ymin,xmax,ymax)
     :return:
     """
+    if len(xyxy) == 0: return xyxy
     xmin, ymin, xmax, ymax = valid_range
     xyxy[:, [0, 2]] = np.clip(xyxy[:, [0, 2]], xmin, xmax)
     xyxy[:, [1, 3]] = np.clip(xyxy[:, [1, 3]], ymin, ymax)
@@ -33,6 +34,7 @@ def clip_cxcywh_minmax(cxcywh, wh_thresh, use_max=True):
     :param use_max: True：最大值限制, False: 最小值限制
     :return:
     """
+    if len(cxcywh) == 0: return cxcywh
     if isinstance(wh_thresh, numbers.Number): wh_thresh = [wh_thresh, wh_thresh]
     if not isinstance(cxcywh, np.ndarray): cxcywh = np.asarray(cxcywh)
     centers = cxcywh.copy()
@@ -51,6 +53,7 @@ def clip_cxcywh_minmax(cxcywh, wh_thresh, use_max=True):
 
 def xyxy2xywh(xyxy: np.ndarray):
     """(xmin,ymin,xmax,ymax)==>(xmin,ymin,w,h)"""
+    if len(xyxy) == 0: return xyxy
     if not isinstance(xyxy, np.ndarray): xyxy = np.asarray(xyxy)
     xywh = xyxy.copy()
     xywh[:, 2] = xywh[:, 2] - xywh[:, 0]  # w=xmax-xmin
@@ -60,6 +63,7 @@ def xyxy2xywh(xyxy: np.ndarray):
 
 def xywh2xyxy(xywh: np.ndarray):
     """(xmin,ymin,w,h)==>(xmin,ymin,xmax,ymax)"""
+    if len(xywh) == 0: return xywh
     if not isinstance(xywh, np.ndarray): xywh = np.asarray(xywh)
     xyxy = xywh.copy()
     xyxy[:, 2] = xyxy[:, 0] + xyxy[:, 2]  # xmax=xmin+w
@@ -69,6 +73,7 @@ def xywh2xyxy(xywh: np.ndarray):
 
 def xyxy2cxcywh(xyxy: np.ndarray, width=None, height=None, normalized=False):
     """(xmin, ymin, xmax, ymax)==>(cx,cy,w,h)"""
+    if len(xyxy) == 0: return xyxy
     if not isinstance(xyxy, np.ndarray): xyxy = np.asarray(xyxy)
     cxcywh = xyxy.copy()
     cxcywh[:, 0] = (xyxy[:, 2] + xyxy[:, 0]) / 2  # cx
@@ -82,6 +87,7 @@ def xyxy2cxcywh(xyxy: np.ndarray, width=None, height=None, normalized=False):
 
 def cxcywh2xyxy(cxcywh: np.ndarray, width=None, height=None, normalized=False):
     """(cx,cy,w,h)==>xmin, ymin, xmax, ymax)"""
+    if len(cxcywh) == 0: return cxcywh
     if not isinstance(cxcywh, np.ndarray): cxcywh = np.asarray(cxcywh)
     xyxy = cxcywh.copy()
     xyxy[:, 0] = cxcywh[:, 0] - cxcywh[:, 2] / 2  # top left x
@@ -103,6 +109,7 @@ def extend_xyxy(xyxy: np.ndarray, scale=[1.0, 1.0], valid_range=[], fixed=False,
                  使用fixed的长宽会按照相同大小扩展
     :return:
     """
+    if len(xyxy) == 0: return xyxy
     if not isinstance(xyxy, np.ndarray): xyxy = np.asarray(xyxy)
     cxcywh = xyxy.copy()
     if fixed:
@@ -130,6 +137,7 @@ def extend_xywh(xywh: np.ndarray, scale=[1.0, 1.0], valid_range=[], fixed=False,
     :param scale: [sx,sy]==>(W,H)
     :return:
     """
+    if len(xywh) == 0: return xywh
     if not isinstance(xywh, np.ndarray): xywh = np.asarray(xywh)
     xyxy = xywh2xyxy(xywh)
     xyxy = extend_xyxy(xyxy, scale, valid_range=valid_range, fixed=fixed, use_max=use_max)
@@ -148,6 +156,7 @@ def extend_xyxy_similar_square(xyxy, use_max=True, weight=1.0, valid_range=[]):
     :param valid_range:有效范围(xmin,ymin,xmax,ymax)
     :return:
     """
+    if len(xyxy) == 0: return xyxy
     if not isinstance(xyxy, np.ndarray): xyxy = np.asarray(xyxy)
     center = xyxy2cxcywh(xyxy)
     if use_max:
@@ -162,6 +171,7 @@ def extend_xyxy_similar_square(xyxy, use_max=True, weight=1.0, valid_range=[]):
     if valid_range: _boxes = clip_xyxy(_boxes, valid_range=valid_range)
     return _boxes
 
+
 def shrink_polygon_pyclipper(polygon, ratio):
     """
     使用Polygon库计算多边形区域的周长和面积，使用pyclipper库进行shrink
@@ -196,6 +206,7 @@ def get_square_boxes(boxes, use_max=True, baseline=-1):
     :param baseline: 当baseline>0，表示正方形最小边长
     :return:
     """
+    if len(boxes) == 0: return boxes
     if not isinstance(boxes, np.ndarray): boxes = np.asarray(boxes)
     center = xyxy2cxcywh(boxes)
     if use_max:
@@ -222,6 +233,7 @@ def get_square_rects(rects, use_max=True, baseline=-1):
     :param baseline: 当baseline>0，表示正方形最小边长
     :return:
     """
+    if len(rects) == 0: return rects
     boxes = xywh2xyxy(rects)
     boxes = get_square_boxes(boxes, use_max=use_max, baseline=baseline)
     rects = xyxy2xywh(boxes)

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/cvutils/video_utils.py

@@ -241,7 +241,7 @@ def video_capture(video_file: int or str, save_video: str or int = None, interva
     video_cap = image_utils.get_video_capture(video_file)
     w, h, num_frames, fps = image_utils.get_video_info(video_cap)
     start = int(kwargs.get("start", 0) * fps)
-    end = int(kwargs.get("end", num_frames / fps) * fps)
+    end = int(kwargs.get("end", num_frames / fps) * fps) if fps > 0 else 0
     end = min(end, num_frames)  # TODO 当num_frames<0时，使用0<end<count继续播放
     interval = fps if interval == -1 else interval  # 当interval=-1，表示interval=fps,即一秒一帧
     save_fps = max(kwargs.get("speed", 1) * fps // interval, 1)
@@ -293,7 +293,7 @@ def video_iterator(video_file: int or str, save_video: str or int = None, interv
     video_cap = image_utils.get_video_capture(video_file)
     w, h, num_frames, fps = image_utils.get_video_info(video_cap)
     start = int(kwargs.get("start", 0) * fps)
-    end = int(kwargs.get("end", num_frames / fps) * fps)
+    end = int(kwargs.get("end", num_frames / fps) * fps) if fps > 0 else 0
     end = min(end, num_frames)  # TODO 当num_frames<0时，使用0<end<count继续播放
     interval = fps if interval == -1 else interval  # 当interval=-1，表示interval=fps,即一秒一帧
     save_fps = max(kwargs.get("speed", 1) * fps // interval, 1)

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/dataloader/parser_coco_det.py

@@ -53,9 +53,9 @@ class CocoDetection(CocoDataset):
 
     def __getitem__(self, index):
         image_id = self.image_ids[index]
-        anno_info, file_info = self.get_object_annotations(image_id)
+        anns_info, file_info = self.get_object_annotations(image_id)
         image, width, height, image_file = self.get_object_image(file_info)
-        boxes, labels = self.get_object_detection(anno_info)
+        boxes, labels = self.get_object_detection(anns_info)
         if self.transform and len(boxes) > 0:
             image, boxes, labels = self.transform(image, boxes, labels)
         num_boxes = len(boxes)
@@ -65,7 +65,9 @@ class CocoDetection(CocoDataset):
         if num_boxes == 0:
             index = int(random.uniform(0, len(self)))
             return self.__getitem__(index)
-        data = {"image": image, "target": target, "labels": labels, "image_id": image_id,
+        data = {"image": image, "boxes": boxes, "labels": labels,
+                "segs": [], "mask": [], "keypoints": [], "target": target,
+                "image_id": image_id, "annotations": anns_info, "file_info": file_info,
                 "image_file": image_file, "size": [width, height], "class_name": self.class_name}
         return data
 

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/dataloader/parser_coco_ins.py

@@ -40,7 +40,8 @@ class CocoInstance(CocoDataset):
         anns_info, file_info = self.get_object_annotations(image_id)
         image, width, height, image_file = self.get_object_image(file_info)
         boxes, labels, mask, segs = self.get_object_instance(anns_info, h=height, w=width, decode=self.decode)
-        data = {"segs": segs, "mask": mask, "image": image, "boxes": boxes, "labels": labels,
+        data = {"image": image, "boxes": boxes, "labels": labels,
+                "segs": segs, "mask": mask, "keypoints": [], "target": [],
                 "image_id": image_id, "annotations": anns_info, "file_info": file_info,
                 "image_file": image_file, "size": [width, height], "class_name": self.class_name}
         return data
@@ -103,7 +104,7 @@ def show_target_image(image, mask, boxes, labels, points=[], class_name=[], thic
     color_image, color_mask = color_utils.draw_image_mask_color(image, mask)
     color_image = image_utils.draw_image_bboxes_labels(color_image, boxes, labels, class_name=class_name,
                                                        thickness=thickness, fontScale=fontScale, drawType="chinese")
-    if len(points)>0:
+    if len(points) > 0:
         color_mask = image_utils.draw_image_contours(color_mask, contours=points)
         color_image = image_utils.draw_image_contours(color_image, contours=points)
     vis_image = image_utils.image_hstack([image, mask, color_image, color_mask])

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/dataloader/parser_coco_kps.py

@@ -32,6 +32,7 @@ class CocoKeypoint(base_coco.CocoDataset):
                                            target_transform=target_transform, use_rgb=use_rgb,
                                            shuffle=shuffle, decode=decode, **kwargs)
         if not class_name: class_name = [self.class_name[1]]
+        if isinstance(class_name, dict): class_name = list(class_name.keys())
         self.kps_info = self.load_keypoints_info(target=class_name)
         self.num_joints = num_joints
         if class_name[0] in BONES:
@@ -70,9 +71,10 @@ class CocoKeypoint(base_coco.CocoDataset):
         anns_info, file_info = self.get_object_annotations(image_id)
         image, width, height, image_file = self.get_object_image(file_info)
         boxes, labels, keypoints = self.get_keypoint_info(anns_info, self.num_joints)
-        data = {"keypoints": keypoints, "image": image, "boxes": boxes, "labels": labels, "image_id": image_id,
-                "annotations": anns_info, "file_info": file_info, "image_file": image_file,
-                "size": [width, height], "class_name": self.class_name}
+        data = {"image": image, "boxes": boxes, "labels": labels,
+                "segs": [], "mask": [], "keypoints": keypoints, "target": [],
+                "image_id": image_id, "annotations": anns_info, "file_info": file_info,
+                "image_file": image_file, "size": [width, height], "class_name": self.class_name}
         return data
 
 
@@ -116,7 +118,7 @@ def CocoKeypoints(anno_file=None,
     return datasets
 
 
-def show_target_image(image, keypoints, boxes, skeleton, colors=None,thickness=2, vis_id=False, use_rgb=True):
+def show_target_image(image, keypoints, boxes, skeleton, colors=None, thickness=2, vis_id=False, use_rgb=True):
     image = image_utils.draw_key_point_in_image(image,
                                                 keypoints,
                                                 pointline=skeleton,

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/dataloader/parser_yolo.py

@@ -217,13 +217,13 @@ class YOLODataset(Dataset):
         image = self.read_image(image_file, use_rgb=self.use_rgb)
         shape = image.shape
         annotation = self.load_annotations(anno_file)
-        boxes, labels, points = self.parser_annotation(annotation, shape, self.class_dict)
+        boxes, labels, points = self.parser_annotation_segs(annotation, shape, self.class_dict)
         data = {"image": image, "boxes": boxes, "labels": labels, "points": points,
                 "image_file": image_file, "anno_file": anno_file}
         return data
 
     @staticmethod
-    def parser_annotation(annotation: dict, shape, class_dict):
+    def parser_annotation_boxes(annotation: dict, shape, class_dict):
         """
         :param annotation:  labelme标注的数据
         :param class_dict:  label映射
@@ -249,6 +249,28 @@ class YOLODataset(Dataset):
             bboxes = image_utils.points2bbox(points)
         return bboxes, labels, points
 
+    @staticmethod
+    def parser_annotation_segs(annotation: dict, shape, class_dict):
+        """
+        :param annotation:  labelme标注的数据
+        :param class_dict:  label映射
+        :param shape: 图片shape(H,W,C),可进行坐标点的维度检查，避免越界
+        :return:
+        """
+        # dim=5,annotation is [class_index, cx, cy, w,  h]
+        # dim=9,annotation is [class_index, x1, y1, x2, y2, x3, y3, x4, y4],四个角点
+        h, w = shape[:2]
+        bboxes, labels, points = [], [], []
+        for anno in annotation:
+            label = anno[0]
+            polys = np.asarray(anno[1:]).reshape(-1, 2)
+            polys = polys * [w, h]
+            boxes = image_utils.polygons2boxes([polys])[0]
+            labels.append(label)
+            points.append(polys)
+            bboxes.append(boxes)
+        return bboxes, labels, points
+
     def index2id(self, index):
         """
         :param index: int or str
@@ -319,16 +341,17 @@ def show_target_image(image, bboxes, labels, points=[], class_name=None, use_rgb
     :param use_rgb:
     :return:
     """
-    image = image_utils.draw_image_contours(image, contours=points)
-    # image = image_utils.draw_image_bboxes_labels(image, bboxes, labels, class_name=class_name)
-    image_utils.cv_show_image("det", image, use_rgb=use_rgb)
+    dst = image.copy()
+    dst = image_utils.draw_image_contours(dst, contours=points, alpha=0.1, thickness=1)
+    dst = image_utils.image_hstack([image, dst])
+    image_utils.cv_show_image("det", dst, use_rgb=use_rgb)
 
 
 if __name__ == "__main__":
     # filename = "/home/dm/nasdata/dataset/csdn/helmet/helmet-dataset-v2/train.txt"
     # filename = "/home/dm/nasdata/dataset/csdn/helmet/helmet-asian/total.txt"
     # filename = "/home/dm/nasdata/dataset/csdn/helmet/helmet-asian/total.txt"
-    filename = "/home/PKing/Downloads/dota8/dota8/images/train/train.txt"
+    filename = "/home/PKing/nasdata/tmp/tmp/medical/肾结石CT/valid/sample.txt"
     dataset = YOLODataset(filename=filename,
                           data_root=None,
                           anno_dir=None,
@@ -341,8 +364,8 @@ if __name__ == "__main__":
     for i in range(len(dataset)):
         print(i)  # i=20
         data = dataset.__getitem__(i)
-        image, bboxes, labels = data["image"], data["box"], data["label"]
-        points = data["point"]
+        image, bboxes, labels = data["image"], data["boxes"], data["labels"]
+        points = data["points"]
         h, w = image.shape[:2]
         image_file = data["image_file"]
         show_target_image(image, bboxes, labels, points=points)

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/file_utils.py

@@ -550,15 +550,15 @@ def move_dir(src, dst, sub=False):
 
 def move_file(srcfile, dstfile):
     """ 移动文件或重命名"""
-    if not os.path.isfile(srcfile):
-        print("%s not exist!" % (srcfile))
-    else:
+    if os.path.exists(srcfile) and os.path.isfile(srcfile):
         fpath, fname = os.path.split(dstfile)  # 分离文件名和路径
         if not os.path.exists(fpath):
             os.makedirs(fpath)  # 创建路径
         shutil.move(srcfile, dstfile)
         # print("copy %s -> %s"%( srcfile,dstfile))
         # time.sleep(1 / 1000.)
+    else:
+        print("%s not exist!" % (srcfile))
 
 
 def copy_file(srcfile, dstfile):

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/image_utils.py

@@ -1905,7 +1905,7 @@ def get_box_crop(image, box):
     return roi
 
 
-get_bbox_crop =get_box_crop
+get_bbox_crop = get_box_crop
 
 
 def get_bboxes_crop(image, bboxes):
@@ -2438,7 +2438,8 @@ def pointPolygonTest(point, contour, measureDist=False):
     return dist
 
 
-def draw_image_contours(image, contours: List[np.ndarray], texts=[], color=(), alpha=0.5, thickness=2, drawType="ch"):
+def draw_image_contours(image, contours: List[np.ndarray], texts=[], color=(), alpha=0.5, thickness=1, fontScale=0.8,
+                        drawType="ch"):
     """
     参考：draw_image_mask_color
     :param image:
@@ -2459,13 +2460,51 @@ def draw_image_contours(image, contours: List[np.ndarray], texts=[], color=(), a
         bgimg = image.copy()
         bgimg = cv2.fillPoly(bgimg, p, color=c)
         image = cv2.addWeighted(src1=image, alpha=1 - alpha, src2=bgimg, beta=alpha, gamma=0)
-        if t: image = draw_text(image, point=(b[0], b[1]), color=c, text=t, thickness=thickness, drawType=drawType)
+        if t: image = draw_text(image, point=(b[0], b[1]), color=c, text=t, thickness=thickness,
+                                fontScale=fontScale, drawType=drawType)
     return image
 
 
 draw_image_mask_color = color_utils.draw_image_mask_color
 
 
+def draw_mask_contours(contours: List[np.ndarray], size, value=255):
+    """
+    参考：draw_image_mask_color
+    :param contours: List[np.ndarray],每个列表是一个轮廓(num_points,1,2)
+    :return:
+    """
+    mask = np.zeros(shape=(size[1], size[0]), dtype=np.uint8)
+    for i in range(0, len(contours)):
+        p = np.asarray(contours[i], dtype=np.int32)
+        if len(p.shape) == 2: p = [p]
+        mask[:] = cv2.drawContours(mask, p, contourIdx=-1, color=value, thickness=-1)
+    return mask
+
+
+def contours_interpolation(contours, n=1000):
+    """
+    对图像轮廓或者对直线进行线性差值，来源于ultralytics/utils/ops.py
+    轮廓插值，线性插值，直线插值
+    Inputs a list of contours (n,2) and returns a list of contours (n,2) up-sampled to n points each.
+
+    Args:
+        contours (list): a list of (n,2) arrays, where n is the number of points in the segment.
+        n (int): number of points to resample the segment to. Defaults to 1000
+
+    Returns:
+        contours (list): the resampled contours.
+    """
+    for i, s in enumerate(contours):
+        s = np.concatenate((s, s[0:1, :]), axis=0)
+        x = np.linspace(0, len(s) - 1, n)
+        xp = np.arange(len(s))
+        contours[i] = (
+            np.concatenate([np.interp(x, xp, s[:, i]) for i in range(2)], dtype=np.float32).reshape(2, -1).T
+        )  # segment xy
+    return contours
+
+
 def get_mask_boundrect_cv(mask, binarize=False, shift=0):
     """
     获得mask的最大外接矩形框(其速度比get_mask_boundrect快2倍左右)
@@ -2903,7 +2942,7 @@ def get_video_capture(video, width=None, height=None, fps=None):
     return video_cap
 
 
-def get_video_info(video_cap: cv2.VideoCapture):
+def get_video_info(video_cap: cv2.VideoCapture, vis=True):
     """
     获得视频的基础信息
     :param video_cap:视频对象
@@ -2913,7 +2952,7 @@ def get_video_info(video_cap: cv2.VideoCapture):
     height = int(video_cap.get(cv2.CAP_PROP_FRAME_HEIGHT))
     num_frames = int(video_cap.get(cv2.CAP_PROP_FRAME_COUNT))
     fps = int(video_cap.get(cv2.CAP_PROP_FPS))
-    print("read video:width:{},height:{},fps:{},num_frames:{}".format(width, height, fps, num_frames))
+    if vis: print("read video:width:{},height:{},fps:{},num_frames:{}".format(width, height, fps, num_frames))
     return width, height, num_frames, fps
 
 

{pybaseutils-2.0.2 → pybaseutils-2.0.5}/pybaseutils/metrics/plot_roc.py

@@ -39,9 +39,9 @@ def plot_roc_curve(true_labels, pred_scores, vis=True):
     # 绘制ROC曲线
     if vis:
         plt_roc_curve([fpr], [tpr], [roc_auc], names=[""])
-        print("FPR           :{}".format(fpr))
-        print("TPR           :{}".format(tpr))
-        print("Threshold     :{}".format(threshold))
+        # print("FPR           :{}".format(fpr))
+        # print("TPR           :{}".format(tpr))
+        # print("Threshold     :{}".format(threshold))
         print("AUC(ROC)      :{}".format(roc_auc))
         print("best_threshold:{} ".format(threshold[optimal_index]))
     return fpr, tpr, roc_auc, threshold, optimal_index
@@ -134,8 +134,60 @@ def custom_roc_curve(y_true, y_score):
     plt.show()
 
 
+def get_best_accuracy(true_labels, pred_scores, threshold=(0, 1.0, 0.05), vis=False):
+    """
+    计算最佳准确率和阈值
+    :param true_labels:
+    :param pred_scores:
+    :param vis:
+    :return: best_acc, best_th
+    """
+    true_labels, pred_scores = np.asarray(true_labels), np.asarray(pred_scores)
+    threshold = np.arange(threshold[0], threshold[1], threshold[2])
+    acc_list = []
+    for th in threshold:
+        pred_labels = np.asarray(pred_scores > th, dtype=np.int32)
+        acc = metrics.accuracy_score(y_true=true_labels, y_pred=pred_labels)
+        acc_list.append(acc)
+    index = np.argmax(acc_list)
+    best_th = threshold[index]
+    best_acc = acc_list[index]
+    if vis:
+        print("best threshold           :{}".format(best_th))
+        print("best Accuracy            :{}".format(best_acc))
+        plt_curve(X=threshold, Y=acc_list, xtitle="threshold", ytitle="Accuracy", title="", grid=True)
+    return best_acc, best_th
+
+
+def plt_curve(X, Y, xtitle="", ytitle="", title="", grid=True):
+    """
+    绘制PR-threshold曲线
+    :param X: Precision
+    :param recall: Recall
+    :param Y
+    :param title:曲线名称
+    :return:
+    """
+    lw = 2
+    plt.figure(figsize=(10, 10))
+    colors = ["b", "r", "c", "m", "g", "y", "k", "w"]
+    plt.plot(X, Y, color=colors[0], lw=lw, label=ytitle)
+    plt.xlim([0.0, 1.05])
+    plt.ylim([0.0, 1.05])
+    # 设置横纵坐标的名称以及对应字体格式
+    font = {'weight': 'normal', 'size': 20}
+    plt.xlabel(xtitle, font)
+    plt.ylabel('', font)
+    plt.title(title, font)
+    plt.legend(loc="lower right")  # "upper right"
+    # plt.legend(loc="upper right")#"upper right"
+    plt.grid(grid)  # 显示网格;
+    plt.show()
+
+
 if __name__ == "__main__":
     true_labels = [0, 0, 0, 0, 0, 1, 1, 1, 1, 1]  # 真实类别
-    pred_scores = [0.1, 0.2, 0.7, 0.3, 0.4, 0.5, 0.2, 0.7, 0.8, 0.9]  # 预测类别分数
+    pred_scores = [0.1, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]  # 预测类别分数
     # pred_scores = [0.11, 0.21, 0.71, 0.31, 0.41, 0.51, 0.21, 0.71, 0.81, 0.91]  # 预测类别分数
-    plot_roc_curve(true_labels, pred_scores)
+    # plot_roc_curve(true_labels, pred_scores)
+    print(get_best_accuracy(true_labels, pred_scores, vis=True))