Cross-Subject划分方式如何删除指定的缺失值？

# -*- coding: utf-8 -*-
import pickle

pkl_file_path = '/root/pyskl/until/no_miss_ntu60_lite-hrnet.pkl'
missing_data_file_path = '/root/pyskl/until/xsub_train_not_in.txt'
val_data_file_path = '/root/pyskl/until/xsub_val_not_in.txt'
output_pkl_file_path = '/root/pyskl/until/exist_miss_ntu60_lite-hrnet.pkl'

# 读取.pkl文件
with open(pkl_file_path, 'rb') as file:
    pkl_data = pickle.load(file)

# print(pkl_data.keys())
# print(type(pkl_data['split']))
# a = pkl_data['split']
# print(a.keys())
# print(type(a['xsub_train']))
# print(len(a['xsub_train']))
# print(type(a['xsub_val']))
# print(len(a['xsub_val']))
print('原始未删除', len(pkl_data['annotations']))

# 读取xsub_train_not_in.txt文件
with open(missing_data_file_path, 'r', encoding='utf-8') as file:
    missing_data = file.read().splitlines()

# 读取xsub_val_not_in.txt文件
with open(val_data_file_path, 'r', encoding='utf-8') as file:
    val_data = file.read().splitlines()

# 获取xsub_train、xsub_val和annotations
xsub_train = pkl_data['split']['xsub_train']
xsub_val = pkl_data['split']['xsub_val']
annotations = pkl_data['annotations']

# 从xsub_train中删除missing_data中的值
for missing_item in missing_data:
    xsub_train = [item for item in xsub_train if item != missing_item]

# 从annotations中删除frame_dir在missing_data中的字典项
for missing_item in missing_data:
    annotations = [annotation for annotation in annotations if annotation['frame_dir'] != missing_item]

# 从xsub_val中删除val_data中的值
for val_item in val_data:
    xsub_val = [item for item in xsub_val if item != val_item]

# 从annotations中删除frame_dir在val_data中的字典项
for val_item in val_data:
    annotations = [annotation for annotation in annotations if annotation['frame_dir'] != val_item]

# 更新pkl_data
pkl_data['split']['xsub_train'] = xsub_train
pkl_data['split']['xsub_val'] = xsub_val
pkl_data['annotations'] = annotations

# 将更新后的数据保存回.pkl文件
with open(output_pkl_file_path, 'wb') as file:
    pickle.dump(pkl_data, file)

print(f"结果已保存到文件: {output_pkl_file_path}")

# 验证更新后的pkl文件
with open(output_pkl_file_path, 'rb') as file:
    updated_pkl_data = pickle.load(file)

print('处理过后的pkl：')
# print(updated_pkl_data.keys())
# print(type(updated_pkl_data['split']))
# a = updated_pkl_data['split']
# print(a.keys())
# print(type(a['xsub_train']))
# print(len(a['xsub_train']))
# print(type(a['xsub_val']))
# print(len(a['xsub_val']))
print('annotation长度', len(updated_pkl_data['annotations']))

把这个xsub_train字段里面删除之前找到的指定的缺失值

把40320这个长度变成40320-229=40091

然后把这个xsub_val字段里面删除之前找到的指定的缺失值

把16560这个长度变成16560-73=16487

而且删除这些缺失值的时候，对应的annotations字段里面出现的frame_dir为对应缺失值的键值也需要删掉

举个例子，假如说frame_dir': 'S005C001P004R002A042对应是缺失值，那就删除整个字典

{'frame_dir': 'S005C001P004R002A042', 'label': 41, 'img_shape': (540, 960), 'original_shape': (540, 960), 'total_frames': 104, 'num_person_raw': 1, 'keypoint': array([[[[600.  , 118.44],
         [605.5 , 113.4 ],
         [595.  , 113.4 ],
         ...,
         [562.  , 280.8 ],
         [577.5 , 329.  ],
         [557.  , 331.5 ]],

        [[600.5 , 118.44],
         [605.5 , 113.4 ],
         [595.5 , 113.4 ],
         ...,
         [562.5 , 280.8 ],
         [577.5 , 329.  ],
         [557.5 , 334.  ]],

        [[601.  , 118.5 ],
         [603.5 , 113.44],
         [596.  , 113.44],
         ...,
         [563.  , 280.8 ],
         [578.  , 329.  ],
         [558.  , 334.  ]],

        ...,

        [[597.  , 128.9 ],
         [602.  , 123.75],
         [591.5 , 123.75],
         ...,
         [565.5 , 289.2 ],
         [540.  , 336.  ],
         [529.5 , 343.8 ]],

        [[599.5 , 128.4 ],
         [605.  , 123.2 ],
         [594.5 , 123.2 ],
         ...,
         [571.  , 289.8 ],
         [540.  , 336.8 ],
         [532.  , 339.2 ]],

        [[603.5 , 127.94],
         [606.  , 122.75],
         [598.5 , 122.75],
         ...,
         [580.  , 286.5 ],
         [536.  , 336.  ],
         [549.  , 338.5 ]]]], dtype=float16), 'keypoint_score': array([[[0.9175, 0.939 , 0.898 , ..., 0.837 , 0.852 , 0.7817],
        [0.9175, 0.9375, 0.896 , ..., 0.836 , 0.848 , 0.7876],
        [0.92  , 0.9336, 0.8936, ..., 0.823 , 0.849 , 0.789 ],
        ...,
        [0.873 , 0.9453, 0.9077, ..., 0.895 , 0.4358, 0.7656],
        [0.8647, 0.9126, 0.892 , ..., 0.885 , 0.5586, 0.6543],
        [0.878 , 0.936 , 0.914 , ..., 0.9053, 0.684 , 0.6074]]],
      dtype=float16)},

这个叫整个字典的意思