[Feature] Add dataset analysis script

tools/analysis_tools/browse_dataset_distribution_function1.py

@@ -0,0 +1,61 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import os.path as osp
+
+import matplotlib.pyplot as plt
+from mmengine.config import Config
+from pycocotools.coco import COCO
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='Distribution of categories and bbox instances')
+    parser.add_argument('config', help='config file path')
+    parser.add_argument('--ann-file', default='train', help='dataset ann-file')
+    parser.add_argument(
+        '--out-dir',
+        default='./data',
+        type=str,
+        help='If there is no display interface, you can save it')
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = parse_args()
+    cfg = Config.fromfile(args.config)
+
+    data_root = cfg.data_root
+    dataset_type = cfg.dataset_type
+    if args.ann_file == 'train':
+        ann_file = cfg.train_dataloader.dataset.ann_file
+    elif args.ann_file == 'val':
+        ann_file = cfg.val_dataloader.dataset.ann_file
+
+    labels = []
+    Num = []
+    coco = COCO(osp.join(data_root, ann_file))
+    categories = coco.loadCats(coco.getCatIds())
+    category_names = [category['name'] for category in categories]
+
+    for category_name in category_names:
+        category = coco.getCatIds(catNms=category_name)
+        annId = coco.getAnnIds(catIds=category)
+        labels.append(category_name)
+        Num.append(len(annId))
+
+    fig = plt.figure()
+    plt.bar(labels, Num, align='center')
+    plt.xticks(rotation=70)
+    # X,Y name and title
+    plt.xlabel('Category')
+    plt.ylabel('Number')
+    plt.title(dataset_type)
+    out = args.out_dir
+    fig.set_size_inches(35, 18)
+    fig.savefig(f'{out}/{dataset_type}.jpg')  # Save Image
+    # plt.show()  # Show Image
+
+
+if __name__ == '__main__':
+    main()

tools/analysis_tools/browse_dataset_distribution_function2.py

@@ -0,0 +1,66 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import os.path as osp
+
+import matplotlib.pyplot as plt
+from mmengine.config import Config
+from pycocotools.coco import COCO
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='Width and height distribution of category and bbox')
+    parser.add_argument('config', help='config file path')
+    parser.add_argument('--ann-file', default='train', help='dataset ann-file')
+    parser.add_argument(
+        '--out-dir',
+        default='./data',
+        type=str,
+        help='If there is no display interface, you can save it')
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = parse_args()
+    cfg = Config.fromfile(args.config)
+
+    data_root = cfg.data_root
+    if args.ann_file == 'train':
+        ann_file = cfg.train_dataloader.dataset.ann_file
+    elif args.ann_file == 'val':
+        ann_file = cfg.val_dataloader.dataset.ann_file
+
+    labels = []
+    ann_bbox_w = []
+    ann_bbow_h = []
+    coco = COCO(osp.join(data_root, ann_file))
+    categories = coco.loadCats(coco.getCatIds())
+    category_names = [category['name'] for category in categories]
+
+    for category_name in category_names:
+        category = coco.getCatIds(catNms=category_name)
+        annId = coco.getAnnIds(catIds=category)
+        labels.append(category_name)
+        annotations = coco.loadAnns(annId)
+        anns = [ann['bbox'] for ann in annotations]
+        for i in range(len(anns)):
+            ann = anns[i]
+            bbox_w = ann[2]
+            bbow_h = ann[3]
+            ann_bbox_w.append(bbox_w)
+            ann_bbow_h.append(bbow_h)
+
+        fig = plt.figure()
+        plt.scatter(ann_bbox_w, ann_bbow_h)
+        plt.xlabel('Width of bbox')
+        plt.ylabel('High of bbox')
+        plt.title(f'Current Display Category:{category_name}')
+        out = args.out_dir
+        fig.set_size_inches(35, 18)
+        plt.savefig(f'{out}/{category_name}.jpg')  # Save Image
+        # plt.show()  # Show Image
+
+
+if __name__ == '__main__':
+    main()

tools/analysis_tools/browse_dataset_distribution_function3.py

@@ -0,0 +1,81 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import os.path as osp
+
+import matplotlib.pyplot as plt
+from mmengine.config import Config
+from pycocotools.coco import COCO
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='Proportional distribution of bbox width and height')
+    parser.add_argument('config', help='config file path')
+    parser.add_argument('--ann-file', default='train', help='dataset ann-file')
+    parser.add_argument(
+        '--out-dir',
+        default='./data',
+        type=str,
+        help='If there is no display interface, you can save it')
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = parse_args()
+    cfg = Config.fromfile(args.config)
+
+    data_root = cfg.data_root
+    if args.ann_file == 'train':
+        ann_file = cfg.train_dataloader.dataset.ann_file
+    elif args.ann_file == 'val':
+        ann_file = cfg.val_dataloader.dataset.ann_file
+
+    labels = []
+    ann_bbox_w = []
+    ann_bbow_h = []
+    ratio = []
+    count_list = list()
+    ratio_sum = []
+    ratio_list = []
+    coco = COCO(osp.join(data_root, ann_file))
+    categories = coco.loadCats(coco.getCatIds())
+    category_names = [category['name'] for category in categories]
+
+    for category_name in category_names:
+        category = coco.getCatIds(catNms=category_name)
+        annId = coco.getAnnIds(catIds=category)
+        labels.append(category_name)
+        annotations = coco.loadAnns(annId)
+        anns = [ann['bbox'] for ann in annotations]
+        for i in range(len(anns)):
+            ann = anns[i]
+            bbox_w = ann[2]
+            bbow_h = ann[3]
+            ann_bbox_w.append(bbox_w)
+            ann_bbow_h.append(bbow_h)
+            ratios = bbox_w / bbow_h
+            ratio.append(round(ratios, 2))
+        count_set = set(ratio)
+        for t in count_set:
+            count_list.append((t, ratio.count(t)))
+        for n in range(len(count_list)):
+            count = count_list[n]
+            ratio_sums = count[1]
+            ratio_lists = count[0]
+            ratio_sum.append(ratio_sums)
+            ratio_list.append(ratio_lists)
+
+        fig = plt.figure()
+        plt.scatter(ratio_list, ratio_sum)
+        plt.xlabel('The width to height ratio')
+        plt.ylabel('Quantity of same width to height ratio')
+        plt.title(category_name)
+        out = args.out_dir
+        fig.set_size_inches(35, 18)
+        plt.savefig(f'{out}/{category_name}.jpg')  # Save Image
+        # plt.show()  # Show Image
+
+
+if __name__ == '__main__':
+    main()

tools/analysis_tools/browse_dataset_distribution_function4.py

@@ -0,0 +1,86 @@
+# Copyright (c) OpenMMLab. All rights reserved.
+import argparse
+import os.path as osp
+
+import matplotlib.pyplot as plt
+from mmengine.config import Config
+from pycocotools.coco import COCO
+
+
+def parse_args():
+    parser = argparse.ArgumentParser(
+        description='Classification and area distribution')
+    parser.add_argument('config', help='train config file path')
+    parser.add_argument('--ann-file', default='train', help='dataset ann-file')
+    parser.add_argument('--size', default=None, help='dataset type')
+    parser.add_argument(
+        '--out-dir',
+        default='./data',
+        type=str,
+        help='If there is no display interface, you can save it')
+
+    args = parser.parse_args()
+    return args
+
+
+def main():
+    args = parse_args()
+    cfg = Config.fromfile(args.config)
+
+    data_root = cfg.data_root
+    if args.ann_file == 'train':
+        ann_file = cfg.train_dataloader.dataset.ann_file
+    elif args.ann_file == 'val':
+        ann_file = cfg.val_dataloader.dataset.ann_file
+
+    labels = []
+    small = []
+    small_num = []
+    medium = []
+    medium_num = []
+    large = []
+    large_num = []
+    coco = COCO(osp.join(data_root, ann_file))
+    categories = coco.loadCats(coco.getCatIds())
+    category_names = [category['name'] for category in categories]
+
+    for category_name in category_names:
+        category = coco.getCatIds(catNms=category_name)
+        annId = coco.getAnnIds(catIds=category)
+        labels.append(category_name)
+        annotations = coco.loadAnns(annId)
+        areas = [area['area'] for area in annotations]
+        for i in range(len(areas)):
+            area = areas[i]
+            if area < 1024:
+                small.append(area)
+            elif 1024 < area < 9216:
+                medium.append(area)
+            else:
+                large.append(area)
+        small_num.append(len(small))
+        medium_num.append(len(medium))
+        large_num.append(len(large))
+    figure_name = args.size
+    fig = plt.figure()
+    if figure_name == 'small':
+        plt.bar(labels, small_num, align='center')
+    elif figure_name == 'medium':
+        plt.bar(labels, medium_num, align='center')
+    elif figure_name == 'large':
+        plt.bar(labels, large_num, align='center')
+    else:
+        print('Please enter the correct size, such as small, medium, large')
+    plt.xticks(rotation=70)
+    # X,Y name and title
+    plt.xlabel('Category')
+    plt.ylabel('Area')
+    plt.title(f'{figure_name}')
+    out = args.out_dir
+    fig.set_size_inches(35, 18)
+    fig.savefig(f'{out}/{figure_name}.jpg')  # Save Image
+    # plt.show()  # Show Image
+
+
+if __name__ == '__main__':
+    main()