pybaseutils 2.2.21__tar.gz → 2.2.24__tar.gz

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.2
 Name: pybaseutils
-Version: 2.2.21
+Version: 2.2.24
 Summary: pybaseutils
 Home-page: https://github.com/PanJinquan/base-utils
 Author: PanJinquan

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/__init__.py

@@ -8,4 +8,4 @@
 # --------------------------------------------------------
 """
 
-__version__ = '2.2.21'
+__version__ = '2.2.24'

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/converter/build_coco.py

@@ -120,7 +120,7 @@ class COCOBuilder():
 
     def save_coco(self, json_file):
         file_utils.create_file_path(json_file)
-        file_utils.write_json_path(json_file, self.coco)
+        file_utils.save_json(json_file, self.coco)
         print("save file:{}".format(json_file))
 
     def set_keypoints_category(self, kps_name=[], skeleton=[], cat_id=0):

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/converter/build_labelme.py

@@ -48,7 +48,7 @@ def maker_labelme(json_file, points, labels, image_name, image_size, image_bs64=
         "imageWidth": image_size[0]
     }
     if os.path.exists(image_name): file_utils.copy_file_to_dir(image_name, os.path.dirname(json_file))
-    file_utils.write_json_path(json_file, data)
+    file_utils.save_json(json_file, data)
     return data
 
 
@@ -60,9 +60,9 @@ def del_labelme_imagedata(anno_dir):
     """
     file_list = file_utils.get_files_lists(anno_dir, postfix=["*.json"])
     for anno_file in tqdm(file_list):
-        data_info = json_utils.read_json_data(anno_file)
+        data_info = json_utils.load_json(anno_file)
         data_info["imageData"] = None
-        json_utils.write_json_path(anno_file, data_info)
+        json_utils.save_json(anno_file, data_info)
 
 
 def copy_labelme_files(image_dir, anno_dir, out_root):
@@ -77,7 +77,7 @@ def copy_labelme_files(image_dir, anno_dir, out_root):
     out_images = file_utils.create_dir(out_root, "images")
     out_json = file_utils.create_dir(out_root, "json")
     for json_file in tqdm(json_list):
-        json_data = json_utils.read_json_data(json_file)
+        json_data = json_utils.load_json(json_file)
         image_name = json_data['imagePath']
         shapes = json_data.get('shapes', [])
         image_file = os.path.join(image_dir, image_name)

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/converter/concat_coco.py

@@ -103,14 +103,14 @@ class ConcatCoco(object):
         """
         if isinstance(file_dict, dict):
             for dirname, file in file_dict.items():
-                coco = file_utils.read_json_data(file)
+                coco = file_utils.load_json(file)
                 self.add_categories(copy.deepcopy(coco["categories"]))
                 self.add_images(copy.deepcopy(coco["images"]), dirname)
                 self.add_annotations(copy.deepcopy(coco["annotations"]),
                                      copy.deepcopy(coco["categories"]))
         elif isinstance(file_dict, list):
             for file in file_dict:
-                coco = file_utils.read_json_data(file)
+                coco = file_utils.load_json(file)
                 self.add_categories(copy.deepcopy(coco["categories"]))
                 self.add_images(copy.deepcopy(coco["images"]), dirname=None)
                 self.add_annotations(copy.deepcopy(coco["annotations"]),
@@ -120,7 +120,7 @@ class ConcatCoco(object):
 
     def save_coco(self, json_file):
         file_utils.create_file_path(json_file)
-        file_utils.write_json_path(json_file, self.coco)
+        file_utils.save_json(json_file, self.coco)
         print("save file:{}".format(json_file))
 
 

pybaseutils-2.2.24/pybaseutils/converter/prelabelme.py

@@ -0,0 +1,115 @@
+# -*- coding: utf-8 -*-
+"""
+# --------------------------------------------------------
+# @Author : Pan
+# @E-mail : 
+# @Date   : 2025-04-29 09:13:09
+# @Brief  :
+# --------------------------------------------------------
+"""
+import os
+import cv2
+from tqdm import tqdm
+from pybaseutils.converter import build_labelme
+from pybaseutils import http_utils, base64_utils, file_utils, image_utils
+
+
+class PreLabelme(object):
+    """
+    调用API接口进行预标注，并保持为Labelme数据格式
+    接口必须返回字段:
+            boxes 必选
+            label 必选
+            score 可选
+            names 可选
+            segms 可选
+    """
+
+    def __init__(self, url, headers=None, timeout=6, max_retries=1):
+        """
+        :param url:
+        """
+        self.url = url
+        self.headers = headers
+        self.timeout = timeout
+        self.max_retries = max_retries
+
+    def request(self, params, **kwargs):
+        """
+        :param params:
+        :return: boxes,label,names,segms,,score
+        """
+        params.update(kwargs)
+        params = base64_utils.serialization(params)
+        result = http_utils.post(self.url, params=params, headers=self.headers,
+                                 timeout=self.timeout, max_retries=self.max_retries)
+        result = base64_utils.deserialization(result)
+        return result
+
+    def draw_result(self, image, result: dict, vis=True, delay=0):
+        """
+        :param image: BGR image
+        :param boxes:<np.ndarray>: (num_boxes, 4), box=[xmin,ymin,xmax,ymax]
+        :param score:<np.ndarray>: (num_boxes,)
+        :param label:<np.ndarray>: (num_boxes,)
+        :return:
+        """
+        color = (255, 0, 0)
+        boxes = result.get("boxes", [])
+        score = result.get("score", [])
+        landm = result.get("landm", [])
+        segms = result.get("segms", [])
+        label = result.get("label", [])
+        names = result.get("names", [])
+        if not names: names = label
+        texts = ["{} {:3.3f}".format(n, s) for n, s in zip(names, score)]
+        image = image_utils.draw_image_bboxes_text(image, boxes, texts, color=color, thickness=2,
+                                                   fontScale=0.8, drawType="ch")
+        image = image_utils.draw_landmark(image, landm, radius=2, fontScale=1.0, color=color)
+        if vis: image_utils.cv_show_image("image", image, delay=delay)
+        return image
+
+    def detect_image_dir(self, image_dir, vis=True):
+        """
+        :param image_dir: directory or image file path
+        :param vis:<bool>
+        :return:
+        """
+        image_list = file_utils.get_files_lists(image_dir)
+        for image_file in tqdm(image_list):
+            image = cv2.imread(image_file)
+            result = self.request(params={"image": image})
+            if vis: image = self.draw_result(image, result)
+
+    def prelabel_labelme(self, image_dir, clsss_dict={}, vis=True):
+        """
+        进行预标注,并输出labelme数据格式
+        :param image_dir: directory or image file path
+        :param vis:<bool>
+        :return:
+        """
+        image_list = file_utils.get_files_lists(image_dir)
+        for image_file in tqdm(image_list):
+            image = cv2.imread(image_file)
+            if image is None: continue
+            h, w = image.shape[:2]
+            result = self.request(params={"image": image})
+            image_name = os.path.basename(image_file)
+            json_file = file_utils.change_postfix(image_file, postfix=".json")
+            boxes = result.get("boxes", [])
+            segms = result.get("segms", [])
+            label = result.get("label", [])
+            names = [clsss_dict[l] for l in label] if clsss_dict else result.get("names", [])
+            result['names'] = names
+            points = image_utils.boxes2polygons(boxes)
+            build_labelme.maker_labelme(json_file, points, names, image_name, image_size=(w, h), image_bs64=None)
+            if vis: image = self.draw_result(image, result)
+
+
+if __name__ == '__main__':
+    # url = 'http://192.168.68.102:5000/detect'
+    url = 'http://0.0.0.0:5000/detect'
+    image_dir = "/media/PKing/新加卷1/个人文件/video/video/frame"
+    s = PreLabelme(url=url)
+    s.prelabel_labelme(image_dir=image_dir, vis=False)
+    # s.detect_image_dir(image_dir=image_dir)

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/coords_utils.py

@@ -23,6 +23,7 @@ def clip_xyxy(xyxy: np.ndarray, valid_range):
     :return:
     """
     if len(xyxy) == 0: return xyxy
+    if not isinstance(xyxy, np.ndarray): xyxy = np.asarray(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)
@@ -105,7 +106,7 @@ def cxcywh2xyxy(cxcywh: np.ndarray, width=None, height=None, normalized=False):
 def extend_xyxy(xyxy: np.ndarray, scale=[1.0, 1.0], valid_range=[], fixed=False, use_max=True):
     """
     :param bboxes: [[xmin, ymin, xmax, ymax]]
-    :param scale: [sx,sy]==>(W,H)
+    :param scale: [sx,sy]==>(W,H),（sx1,sy1,sx2,sy2）,1.0表示不缩放，<1.0表示缩小倍数，>1.0表示扩大倍数
     :param valid_range:有效范围(xmin,ymin,xmax,ymax)
     :param fixed: 长宽是否按照相同大小扩展
                  当长宽比相差比较大，直接scale会导致短边扩展不明显，
@@ -114,22 +115,30 @@ def extend_xyxy(xyxy: np.ndarray, scale=[1.0, 1.0], valid_range=[], fixed=False,
     """
     if len(xyxy) == 0 or len(scale) == 0: return xyxy
     if not isinstance(xyxy, np.ndarray): xyxy = np.asarray(xyxy)
-    cxcywh = xyxy.copy()
-    if fixed:
-        cxcywh[:, 0] = (xyxy[:, 2] + xyxy[:, 0]) / 2  # cx
-        cxcywh[:, 1] = (xyxy[:, 3] + xyxy[:, 1]) / 2  # cy
-        dw = (xyxy[:, 2] - xyxy[:, 0]) * (scale[0] - 1)
-        dh = (xyxy[:, 3] - xyxy[:, 1]) * (scale[1] - 1)
-        dp = np.vstack((dw, dh))
-        dp = np.max(dp, axis=0) if use_max else np.min(dp, axis=0)
-        cxcywh[:, 2] = (xyxy[:, 2] - xyxy[:, 0]) + dp  # w
-        cxcywh[:, 3] = (xyxy[:, 3] - xyxy[:, 1]) + dp  # h
-    else:
-        cxcywh[:, 0] = (xyxy[:, 2] + xyxy[:, 0]) / 2  # cx
-        cxcywh[:, 1] = (xyxy[:, 3] + xyxy[:, 1]) / 2  # cy
-        cxcywh[:, 2] = (xyxy[:, 2] - xyxy[:, 0]) * scale[0]  # w
-        cxcywh[:, 3] = (xyxy[:, 3] - xyxy[:, 1]) * scale[1]  # h
-    dxyxy = cxcywh2xyxy(cxcywh, width=None, height=None, normalized=False)
+    if len(scale) == 2:
+        cxcywh = xyxy.copy()
+        if fixed:
+            cxcywh[:, 0] = (xyxy[:, 2] + xyxy[:, 0]) / 2  # cx
+            cxcywh[:, 1] = (xyxy[:, 3] + xyxy[:, 1]) / 2  # cy
+            dw = (xyxy[:, 2] - xyxy[:, 0]) * (scale[0] - 1)
+            dh = (xyxy[:, 3] - xyxy[:, 1]) * (scale[1] - 1)
+            dp = np.vstack((dw, dh))
+            dp = np.max(dp, axis=0) if use_max else np.min(dp, axis=0)
+            cxcywh[:, 2] = (xyxy[:, 2] - xyxy[:, 0]) + dp  # w
+            cxcywh[:, 3] = (xyxy[:, 3] - xyxy[:, 1]) + dp  # h
+        else:
+            cxcywh[:, 0] = (xyxy[:, 2] + xyxy[:, 0]) / 2  # cx
+            cxcywh[:, 1] = (xyxy[:, 3] + xyxy[:, 1]) / 2  # cy
+            cxcywh[:, 2] = (xyxy[:, 2] - xyxy[:, 0]) * scale[0]  # w
+            cxcywh[:, 3] = (xyxy[:, 3] - xyxy[:, 1]) * scale[1]  # h
+        dxyxy = cxcywh2xyxy(cxcywh, width=None, height=None, normalized=False)
+    elif len(scale) == 4:
+        xywh = xyxy2xywh(xyxy)
+        xyxy[:, 0] = xyxy[:, 0] + xywh[:, 2] * (1 - scale[0])
+        xyxy[:, 1] = xyxy[:, 1] + xywh[:, 3] * (1 - scale[1])
+        xyxy[:, 2] = xyxy[:, 2] - xywh[:, 2] * (1 - scale[2])
+        xyxy[:, 3] = xyxy[:, 3] - xywh[:, 3] * (1 - scale[3])
+        dxyxy = xyxy
     if valid_range: dxyxy = clip_xyxy(dxyxy, valid_range=valid_range)
     return dxyxy
 

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/cvutils/corner_utils.py

@@ -76,7 +76,7 @@ def get_target_points(src_pts: np.ndarray):
     return dst_pts
 
 
-def get_order_points(src_pts):
+def get_order_points_index(src_pts):
     """
     对4个点按顺时针方向进行排序:[top-left, top-right, bottom-right, bottom-left]
     top-left    ：对应y+x之和的最小点
@@ -91,13 +91,32 @@ def get_order_points(src_pts):
     :param src_pts: pts_dst [top-left, top-right, bottom-right, bottom-left]
     :return:
     """
-    dst_pts = np.zeros(shape=(4, 2), dtype=np.float32)
     s = src_pts.sum(axis=1)
-    dst_pts[0] = src_pts[np.argmin(s)]
-    dst_pts[2] = src_pts[np.argmax(s)]
     d = np.diff(src_pts, axis=1)
-    dst_pts[1] = src_pts[np.argmin(d)]
-    dst_pts[3] = src_pts[np.argmax(d)]
+    i0 = np.argmin(s)
+    i2 = np.argmax(s)
+    i1 = np.argmin(d)
+    i3 = np.argmax(d)
+    return [i0, i1, i2, i3]
+
+
+def get_order_points(src_pts):
+    """
+    对4个点按顺时针方向进行排序:[top-left, top-right, bottom-right, bottom-left]
+    top-left    ：对应y+x之和的最小点
+    bottom-right：对应y+x之和的最大点
+    top-right   ：对应y-x之差的最小点
+    bottom-left ：对应y-x之差的最大点
+         0(top-left)----(w10)----1(top-right)
+            |                       |
+          (h30)                    (h21)
+            |                       |
+        3(bottom-left)--(w23)---2(bottom-right)
+    :param src_pts: pts_dst [top-left, top-right, bottom-right, bottom-left]
+    :return:
+    """
+    index = get_order_points_index(src_pts)
+    dst_pts = src_pts[index]
     return dst_pts
 
 

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/cvutils/video_utils.py

@@ -61,7 +61,7 @@ def video2gif(video_file, gif_file=None, func=None, interval=1, use_pil=False, f
         image_utils.frames2gif_by_imageio(frames, gif_file=gif_file, fps=fps, loop=0)
 
 
-def video2frames(video_file, out_dir=None, func=None, interval=1, vis=True, delay=10):
+def video2frames(video_file, out_dir=None, func=None, interval=1, start=0, end=-1, vis=True, delay=10):
     """
     视频抽帧图像
     :param video_file: 视频文件
@@ -73,26 +73,14 @@ def video2frames(video_file, out_dir=None, func=None, interval=1, vis=True, dela
     """
     name = os.path.basename(video_file).split(".")[0]
     if not out_dir:  out_dir = os.path.join(os.path.dirname(video_file), name)
-    video_cap = get_video_capture(video_file)
-    width, height, num_frames, fps = get_video_info(video_cap)
-    if not interval: interval = fps
-    if not os.path.exists(out_dir): os.makedirs(out_dir)
-    count = 0
-    while True:
-        if count % interval == 0:
-            # 设置抽帧的位置
-            video_cap.set(cv2.CAP_PROP_POS_FRAMES, count)
-            isSuccess, frame = video_cap.read()
-            if not isSuccess or 0 < num_frames < count: break
-            if func:
-                frame = func(frame)
-            if vis:
-                image_utils.cv_show_image("frame", frame, use_rgb=False, delay=delay)
-            frame_file = os.path.join(out_dir, "{}_{:0=4d}.jpg".format(name, count))
-            cv2.imwrite(frame_file, frame)
-        count += 1
-    video_cap.release()
-    cv2.destroyAllWindows()
+    video_cap = video_iterator(video_file, save_video=None, interval=interval, start=start, end=end)
+    for data_info in video_cap:
+        frame = data_info["frame"]
+        count = data_info["count"]
+        if func: frame = func(frame)
+        if vis: image_utils.cv_show_image("frame", frame, use_rgb=False, delay=delay)
+        frame_file = os.path.join(out_dir, "{}_{:0=4d}.jpg".format(name, count))
+        cv2.imwrite(frame_file, frame)
 
 
 def video2frames_similarity(video_file, out_dir=None, func=None, interval=1, thresh=0.3, vis=True):
@@ -238,6 +226,7 @@ def video_capture(video_file: int or str, save_video: str or int = None, interva
                  title: 控制显示窗口名
     :return:
     """
+    assert os.path.exists(video_file), f"video_file={video_file}"
     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)
@@ -291,11 +280,12 @@ def video_iterator(video_file: int or str, save_video: str or int = None, interv
                  speed: 播放速度
     :return: frame, count, w, h, fps =data_info['frame'],data_info['count'],data_info['w'],data_info['h'],data_info['fps']
     """
+    assert os.path.exists(video_file), f"video_file={video_file}"
     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) 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
+    end = min(end, num_frames) if end > 0 else 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)
     count = 0

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/dataloader/base_dataset.py

@@ -129,23 +129,16 @@ class Dataset(object):
         return Dataset.read_file(filename, split=split)
 
     @staticmethod
-    def get_file_list(dir_root, postfix=['*.jpg'], basename=True):
+    def get_file_list(dir_root, postfix=['*.jpg'], sub=False, basename=True):
         """
         获得文件列表
         :param dir_root: 图片文件目录
         :param postfix: 后缀名，可是多个如，['*.jpg','*.png']
+        :param sub: 是否去除根路径
         :param basename: 返回的列表是文件名（True），还是文件的完整路径(False)
         :return:
         """
-        file_list = []
-        for format in postfix:
-            format = os.path.join(dir_root, format)
-            image_list = glob.glob(format)
-            if image_list:
-                file_list += image_list
-        file_list = sorted(file_list)
-        if basename:
-            file_list = [os.path.basename(f).split(".")[0] for f in file_list]
+        file_list = file_utils.get_files_list(dir_root, prefix="", postfix=postfix, sub=sub, basename=basename)
         return file_list
 
     @staticmethod

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/dataloader/parser_image_text.py

@@ -47,6 +47,7 @@ class TextDataset(Dataset):
         self.check = check
         self.kwargs = kwargs
         self.label_index = kwargs.get("label_index", "label")  # 类别字段key
+        self.crop_scale = kwargs.get("crop_scale", [])  # TODO bbox缩放系数
         self.class_name, self.class_dict = self.parser_classes(class_name)
         self.item_list = self.parser_dataset(data_file, data_root=data_root, label_index=self.label_index,
                                              shuffle=shuffle, check=check)
@@ -68,6 +69,7 @@ class TextDataset(Dataset):
 
     def info(self, save_info=""):
         print("----------------------- {} DATASET INFO -----------------------".format(self.phase.upper()))
+        print("Dataset kwargs        :{}".format(self.kwargs))
         print("Dataset num_samples   :{}".format(len(self.item_list)))
         print("Dataset num_classes   :{}".format(self.num_classes))
         print("Dataset class_name    :{}".format(self.class_name))
@@ -81,10 +83,10 @@ class TextDataset(Dataset):
             class_lack = sorted(class_lack.items(), key=lambda x: x[1], reverse=True)
             class_lack = {n[0]: n[1] for n in class_lack}
             class_lack.update({"mean": m})
-            file_utils.write_json_path(os.path.join(save_info, f"{self.phase}_class_dict.json"), self.class_dict)
-            file_utils.write_json_path(os.path.join(save_info, f"{self.phase}_class_count.json"), self.class_count)
+            file_utils.save_json(os.path.join(save_info, f"{self.phase}_class_dict.json"), self.class_dict)
+            file_utils.save_json(os.path.join(save_info, f"{self.phase}_class_count.json"), self.class_count)
             file_utils.write_list_data(os.path.join(save_info, f"{self.phase}_class_name.txt"), self.class_name)
-            file_utils.write_json_path(os.path.join(save_info, f"{self.phase}_class_lack.json"), class_lack)
+            file_utils.save_json(os.path.join(save_info, f"{self.phase}_class_lack.json"), class_lack)
             print("loss_labels: {}".format(class_lack))
         print("------------------------------------------------------------------")
 
@@ -163,10 +165,9 @@ class TextDataset(Dataset):
         file, label, bbox = item["file"], item[self.label_index], item.get("bbox", [])
         name = self.class_name[label]
         image = self.read_image(file, use_rgb=self.use_rgb)
-        image = self.crop_image(image, bbox=bbox, **self.kwargs) if bbox else image
+        image = self.crop_image(image, bbox=bbox, **self.kwargs)
         if self.transform:
-            image = Image.fromarray(image)
-            image = self.transform(image)
+            image = self.transform(Image.fromarray(image))
         if image is None:
             index = int(random.uniform(0, self.num_samples))
             return self.__getitem__(index)
@@ -178,7 +179,13 @@ class TextDataset(Dataset):
         return len(self.item_list)
 
     def crop_image(self, image, bbox, **kwargs):
-        """裁剪图片"""
+        """
+        裁剪图片
+        :param image:
+        :param bbox:
+        :param kwargs:  use_max,use_mean,crop_scale
+        :return:
+        """
         if len(bbox) == 0: return image
         boxes = image_utils.get_square_boxes(boxes=[bbox],
                                              use_max=kwargs.get("use_max", False),

{pybaseutils-2.2.21 → pybaseutils-2.2.24}/pybaseutils/dataloader/parser_labelme.py

@@ -150,11 +150,11 @@ class LabelMeDataset(Dataset):
         if not image_dir:
             image_dir = self.search_path(data_root, ["images", "JPEGImages"])
         if image_dir and not image_ids:
-            image_ids = self.get_file_list(image_dir, postfix=file_utils.IMG_POSTFIX, basename=False)
-            image_ids = [os.path.basename(f) for f in image_ids]
+            image_ids = self.get_file_list(image_dir, postfix=file_utils.IMG_POSTFIX, sub=True, basename=False)
+            if not anno_dir: anno_dir = image_dir
         elif anno_dir and not image_ids:
-            image_ids = self.get_file_list(anno_dir, postfix=["*.json", "*.xml"], basename=False)
-            image_ids = [os.path.basename(f) for f in image_ids]
+            image_ids = self.get_file_list(anno_dir, postfix=file_utils.IMG_POSTFIX, sub=True, basename=False)
+            if not image_dir: image_dir = anno_dir
         assert isinstance(anno_dir, str) and os.path.exists(anno_dir), "no anno_dir :{}".format(anno_dir)
         assert isinstance(image_dir, str) and os.path.exists(image_dir), "no image_dir:{}".format(image_dir)
         assert len(image_ids) > 0, f"image_ids is empty,image_dir={image_dir},anno_dir={anno_dir}"
@@ -330,7 +330,7 @@ class LabelMeDataset(Dataset):
             width = annotation.get('imageWidth', -1)
             height = annotation.get('imageHeight', -1)
         except:
-            print("illegal annotation:{}".format(anno_file))
+            # print("illegal annotation:{}".format(anno_file))
             annos = []
             width = -1
             height = -1
@@ -487,8 +487,7 @@ def show_target_image(image, boxes, labels, points, keypoints=[], color=(), thic
 if __name__ == "__main__":
     from pybaseutils.converter import build_labelme
 
-    anno_dir = "/home/PKing/Downloads/sample/images"
-    # anno_dir = "/home/PKing/Downloads/冲击试验/sample/images"
+    anno_dir = "/home/PKing/nasdata/dataset-dmai/AIJE/dataset/aije-action-cvlm-v2/train-v2/01-核相操作/dataset-v01/images"
     names = None
     dataset = LabelMeDatasets(filename=None,
                               data_root=None,

pybaseutils-2.2.24/pybaseutils/dataloader/parser_labelme_crop.py

@@ -0,0 +1,136 @@
+# -*- coding: utf-8 -*-
+"""
+# --------------------------------------------------------
+# @Author : panjq
+# @Date   : 2020-9-20 13:18:34
+# --------------------------------------------------------
+"""
+
+import os
+import PIL.Image as Image
+import numpy as np
+import random
+import cv2
+import torch
+import copy
+from tqdm import tqdm
+from typing import Dict, List
+from pybaseutils import image_utils, file_utils
+from pybaseutils.dataloader import parser_labelme, parser_image_text
+
+
+class LabelmeDataset(parser_image_text.TextDataset):
+    """加载Labelme数据并进行裁剪,图片和json文件都保存在images目录下"""
+
+    def __init__(self, filename, class_name=None, transform=None, use_rgb=False, shuffle=False,
+                 phase="test", disp=False, check=False, **kwargs):
+        """
+        :param filename:
+        :param class_name:
+        :param transform:
+        :param use_rgb:
+        :param shuffle:
+        :param phase:
+        :param disp:
+        :param check:
+        :param kwargs:  use_max,use_mean,crop_scale
+        """
+        self.dataset = None
+        super(LabelmeDataset, self).__init__(data_file=filename,
+                                             data_root=None,
+                                             class_name=class_name,
+                                             transform=transform,
+                                             shuffle=shuffle,
+                                             use_rgb=use_rgb,
+                                             phase=phase,
+                                             disp=disp,
+                                             check=check,
+                                             label_index="label",
+                                             **kwargs)
+        print("LabelmeDataset    have images:{},have samples:{}".format(len(self.item_list), self.num_samples))
+
+    def parser_dataset(self, data_file, **kwargs):
+        """
+        获得Labelme所有目标信息
+        :param dataset:
+        :return:
+        """
+        self.dataset = parser_labelme.LabelMeDatasets(filename=None,
+                                                      data_root=None,
+                                                      anno_dir=data_file,
+                                                      image_dir=None,
+                                                      class_name=self.class_name,
+                                                      check=False,
+                                                      phase=self.phase,
+                                                      use_rgb=self.use_rgb,
+                                                      shuffle=self.shuffle,
+                                                      read_image=False)
+        print("Please wait, it's in checking")
+        item_list = []
+        for index in tqdm(range(len(self.dataset))):
+            info = self.dataset.__getitem__(index)
+            file, boxes = info["image_file"], info.get("boxes", [])
+            for i in range(len(boxes)):
+                if len(boxes[i]) == 0: continue
+                item_list.append(dict(file=file,
+                                      bbox=boxes[i],
+                                      label=info["labels"][i],
+                                      name=info["names"][i],
+                                      point=info["points"][i],
+                                      )
+                                 )
+        assert len(item_list) > 0, f"item_list is empty, check your data_file is ={data_file}"
+        return item_list
+
+    def __getitem__(self, index):
+        """
+        :param index:
+        :return: {"image": image, "label": label}
+        """
+        item: dict = copy.deepcopy(self.item_list[index])  # TODO Fix a bug: 避免修改原始数据
+        file, bbox = item["file"], item.get("bbox", [])
+        image = self.read_image(file, use_rgb=self.use_rgb)
+        image = self.crop_image(image, bbox=bbox, **self.kwargs)
+        if self.transform:
+            image = self.transform(Image.fromarray(image))
+        if image is None:
+            index = int(random.uniform(0, len(self)))
+            return self.__getitem__(index)
+        return dict(file=file, image=image, label=item["label"], name=item["name"])
+
+
+if __name__ == '__main__':
+    from torchvision import transforms
+
+    filename = "/home/PKing/nasdata/tmp/tmp/RealFakeFace/face-gesture/dataset-v3/image"
+    # filename = "/home/PKing/nasdata/tmp/tmp/RealFakeFace/living/dataset-v2/image"
+    batch_size = 1
+    crop_scale = (1.2, 1.2)
+    input_size = [224, 224]
+    rgb_mean = [0., 0., 0.]
+    rgb_std = [1.0, 1.0, 1.0]
+    class_name = ["face", "闭眼", "张嘴", "低头", "侧脸"]
+    transform = transforms.Compose([
+        transforms.Resize(input_size),
+        transforms.ToTensor(),
+        transforms.Normalize(mean=rgb_mean, std=rgb_std),
+    ])
+
+    dataset = LabelmeDataset(filename=filename,
+                             transform=transform,
+                             resample=True,
+                             shuffle=False,
+                             class_name=class_name,
+                             use_rgb=False,
+                             disp=True,
+                             crop_scale=crop_scale,
+                             vis=False)
+    for i in range(len(dataset)):
+        data_info = dataset.__getitem__(i)
+        file = data_info["file"]
+        image, label, name = data_info["image"], data_info["label"], data_info["name"]
+        image = np.asarray(image).transpose(1, 2, 0)  # 通道由[c,h,w]->[h,w,c]
+        image = np.asarray(image * 255, dtype=np.uint8)
+        label = np.asarray(label, dtype=np.int32)
+        print("{},image.shape:{},label:{},name:{}".format(file, image.shape, label, name))
+        image_utils.cv_show_image("image", image)