python - 将字典映射到向量以获得一组索引向量

Question

我有以下长度为 200 的向量，其中包含引用剪辑列表，如下所示:

clips_reference_name=['v_ApplyEyeMakeup_g08_c01',
 'v_ApplyEyeMakeup_g08_c02',
 'v_ApplyEyeMakeup_g08_c03',
 'v_ApplyEyeMakeup_g08_c04',
 'v_ApplyEyeMakeup_g08_c05',
 'v_ApplyEyeMakeup_g09_c01',
 'v_ApplyEyeMakeup_g09_c02',
 'v_ApplyEyeMakeup_g09_c03',
 'v_ApplyEyeMakeup_g09_c04',
 'v_ApplyEyeMakeup_g09_c05',
 'v_ApplyEyeMakeup_g09_c06',
 'v_ApplyEyeMakeup_g09_c07',
 'v_ApplyEyeMakeup_g10_c01',
 'v_ApplyEyeMakeup_g10_c02',
 'v_ApplyEyeMakeup_g10_c03',
 'v_ApplyEyeMakeup_g10_c04',
 'v_ApplyEyeMakeup_g10_c05',
 'v_ApplyEyeMakeup_g11_c01',
 'v_ApplyEyeMakeup_g11_c02',
 'v_ApplyEyeMakeup_g11_c03',
 'v_ApplyLipstick_g08_c01',
 'v_ApplyLipstick_g08_c02',
 'v_ApplyLipstick_g08_c03',
 'v_ApplyLipstick_g08_c04',
 'v_ApplyLipstick_g09_c01',
 'v_ApplyLipstick_g09_c02',
 'v_ApplyLipstick_g09_c03',
 'v_ApplyLipstick_g09_c04',
 'v_ApplyLipstick_g10_c01',
 'v_ApplyLipstick_g10_c02',
 'v_ApplyLipstick_g10_c03',
 'v_ApplyLipstick_g10_c04',
 'v_ApplyLipstick_g11_c01',
 'v_ApplyLipstick_g11_c02',
 'v_ApplyLipstick_g11_c03',
 'v_ApplyLipstick_g11_c04',
 'v_ApplyLipstick_g12_c01',
 'v_ApplyLipstick_g12_c02',
 'v_ApplyLipstick_g12_c03',
 'v_ApplyLipstick_g12_c04',
 'v_Archery_g08_c01',
 'v_Archery_g08_c02',
 'v_Archery_g08_c03',
 'v_Archery_g08_c04',
 'v_Archery_g08_c05',
 'v_Archery_g09_c01',
 'v_Archery_g09_c02',
 'v_Archery_g09_c03',
 'v_Archery_g09_c04',
 'v_Archery_g09_c05',
 'v_Archery_g09_c06',
 'v_Archery_g09_c07',
 'v_Archery_g10_c01',
 'v_Archery_g10_c02',
 'v_Archery_g10_c03',
 'v_Archery_g10_c04',
 'v_Archery_g10_c05',
 'v_Archery_g10_c06',
 'v_Archery_g10_c07',
 'v_Archery_g11_c01',
 'v_BabyCrawling_g08_c01',
 'v_BabyCrawling_g08_c02',
 'v_BabyCrawling_g08_c03',
 'v_BabyCrawling_g08_c04',
 'v_BabyCrawling_g09_c01',
 'v_BabyCrawling_g09_c02',
 'v_BabyCrawling_g09_c03',
 'v_BabyCrawling_g09_c04',
 'v_BabyCrawling_g09_c05',
 'v_BabyCrawling_g09_c06',
 'v_BabyCrawling_g10_c01',
 'v_BabyCrawling_g10_c02',
 'v_BabyCrawling_g10_c03',
 'v_BabyCrawling_g10_c04',
 'v_BabyCrawling_g10_c05',
 'v_BabyCrawling_g11_c01',
 'v_BabyCrawling_g11_c02',
 'v_BabyCrawling_g11_c03',
 'v_BabyCrawling_g11_c04',
 'v_BabyCrawling_g12_c01',
 'v_BalanceBeam_g08_c01',
 'v_BalanceBeam_g08_c02',
 'v_BalanceBeam_g08_c03',
 'v_BalanceBeam_g08_c04',
 'v_BalanceBeam_g09_c01',
 'v_BalanceBeam_g09_c02',
 'v_BalanceBeam_g09_c03',
 'v_BalanceBeam_g09_c04',
 'v_BalanceBeam_g10_c01',
 'v_BalanceBeam_g10_c02',
 'v_BalanceBeam_g10_c03',
 'v_BalanceBeam_g10_c04',
 'v_BalanceBeam_g11_c01',
 'v_BalanceBeam_g11_c02',
 'v_BalanceBeam_g11_c03',
 'v_BalanceBeam_g11_c04',
 'v_BalanceBeam_g12_c01',
 'v_BalanceBeam_g12_c02',
 'v_BalanceBeam_g12_c03',
 'v_BandMarching_g08_c01',
 'v_BandMarching_g08_c02',
 'v_BandMarching_g08_c03',
 'v_BandMarching_g08_c04',
 'v_BandMarching_g08_c05',
 'v_BandMarching_g08_c06',
 'v_BandMarching_g08_c07',
 'v_BandMarching_g09_c01',
 'v_BandMarching_g09_c02',
 'v_BandMarching_g09_c03',
 'v_BandMarching_g09_c04',
 'v_BandMarching_g09_c05',
 'v_BandMarching_g09_c06',
 'v_BandMarching_g09_c07',
 'v_BandMarching_g10_c01',
 'v_BandMarching_g10_c02',
 'v_BandMarching_g10_c03',
 'v_BandMarching_g10_c04',
 'v_BandMarching_g10_c05',
 'v_BandMarching_g10_c06',
 'v_BandMarching_g10_c07',
 'v_BaseballPitch_g08_c01',
 'v_BaseballPitch_g08_c02',
 'v_BaseballPitch_g08_c03',
 'v_BaseballPitch_g08_c04',
 'v_BaseballPitch_g08_c05',
 'v_BaseballPitch_g08_c06',
 'v_BaseballPitch_g08_c07',
 'v_BaseballPitch_g09_c01',
 'v_BaseballPitch_g09_c02',
 'v_BaseballPitch_g09_c03',
 'v_BaseballPitch_g09_c04',
 'v_BaseballPitch_g09_c05',
 'v_BaseballPitch_g09_c06',
 'v_BaseballPitch_g09_c07',
 'v_BaseballPitch_g10_c01',
 'v_BaseballPitch_g10_c02',
 'v_BaseballPitch_g10_c03',
 'v_BaseballPitch_g10_c04',
 'v_BaseballPitch_g10_c05',
 'v_BaseballPitch_g11_c01',
 'v_Basketball_g08_c01',
 'v_Basketball_g08_c02',
 'v_Basketball_g08_c03',
 'v_Basketball_g08_c04',
 'v_Basketball_g09_c01',
 'v_Basketball_g09_c02',
 'v_Basketball_g09_c03',
 'v_Basketball_g09_c04',
 'v_Basketball_g09_c05',
 'v_Basketball_g10_c01',
 'v_Basketball_g10_c02',
 'v_Basketball_g10_c03',
 'v_Basketball_g10_c04',
 'v_Basketball_g10_c05',
 'v_Basketball_g11_c01',
 'v_Basketball_g11_c02',
 'v_Basketball_g11_c03',
 'v_Basketball_g11_c04',
 'v_Basketball_g11_c05',
 'v_Basketball_g12_c01',
 'v_BasketballDunk_g08_c01',
 'v_BasketballDunk_g08_c02',
 'v_BasketballDunk_g08_c03',
 'v_BasketballDunk_g08_c04',
 'v_BasketballDunk_g08_c05',
 'v_BasketballDunk_g09_c01',
 'v_BasketballDunk_g09_c02',
 'v_BasketballDunk_g09_c03',
 'v_BasketballDunk_g09_c04',
 'v_BasketballDunk_g09_c05',
 'v_BasketballDunk_g10_c01',
 'v_BasketballDunk_g10_c02',
 'v_BasketballDunk_g10_c03',
 'v_BasketballDunk_g10_c04',
 'v_BasketballDunk_g10_c05',
 'v_BasketballDunk_g11_c01',
 'v_BasketballDunk_g11_c02',
 'v_BasketballDunk_g11_c03',
 'v_BasketballDunk_g11_c04',
 'v_BasketballDunk_g11_c05',
 'v_BenchPress_g08_c01',
 'v_BenchPress_g08_c02',
 'v_BenchPress_g08_c03',
 'v_BenchPress_g08_c04',
 'v_BenchPress_g08_c05',
 'v_BenchPress_g08_c06',
 'v_BenchPress_g08_c07',
 'v_BenchPress_g09_c01',
 'v_BenchPress_g09_c02',
 'v_BenchPress_g09_c03',
 'v_BenchPress_g09_c04',
 'v_BenchPress_g09_c05',
 'v_BenchPress_g09_c06',
 'v_BenchPress_g09_c07',
 'v_BenchPress_g10_c01',
 'v_BenchPress_g10_c02',
 'v_BenchPress_g10_c03',
 'v_BenchPress_g10_c04',
 'v_BenchPress_g11_c01',
 'v_BenchPress_g11_c02']

每个剪辑引用名称都与一组图像相关联。例如:clips_reference_name 中的第一个引用。 'v_ApplyEyeMakeup_g08_c01'，是与一组图像(本例中为 300 个图像)相关联，在以下代码中称为 labels:

v_ApplyEyeMakeup_g08_c01.**0001**.jpeg, ..., v_ApplyEyeMakeup_g08_c01.**0300**.jpeg,

每个引用名称的图像数量因图像而异。

我有一个帧字典(图像名称)，它们的值如下:

dataset= dict(zip(labels, frames))

其中 labels 是一个包含如下值的列表:

v_BasketballDunk_g08_c04_0018.jpeg
v_BandMarching_g10_c05_0097.jpeg
v_BabyCrawling_g11_c01_0010.jpeg
v_ApplyEyeMakeup_g09_c04_0148.jpeg
v_Archery_g08_c01_0008.jpeg
v_BalanceBeam_g11_c02_0058.jpeg
v_BaseballPitch_g09_c05_0002.jpeg
v_ApplyLipstick_g08_c02_0044.jpeg
v_Basketball_g11_c01_0062.jpeg
v_BenchPress_g11_c02_0012.jpeg

帧是 2048 个值的一维向量。

例如:从(标签，框架)创建的字典的第一项如下:

{'v_BasketballDunk_g08_c02_0053.jpeg':
    array([ 0.88717347,  0.51302141,  0.87405699, ...,  0.41013849,
            0.38836521,  0.37444678], dtype=float32), .....}

我想要得到什么？

由于我在 clips_reference_name 中有 200 个项目，因此我想获得与每个项目相对应的 200 个向量，如下所示:

vector-labels_v_ApplyEyeMakeup_g08_c02 = [v_ApplyEyeMakeup_g08_c02_0001.jpeg,
                                          v_ApplyEyeMakeup_g08_c02_0002.jpeg ,
                                          ...,
                                          v_ApplyEyeMakeup_g08_c02_0300.jpeg]
 vector-frme-values_v_ApplyEyeMakeup_g08_c02 = [[0.47,...,0.98], ..., [0.17,...,0.45]]

 vector_labels-v_BabyCrawling_g09_c02 = [v_BabyCrawling_g09_c02_0001.jpeg,
                                         v_BabyCrawling_g09_c02_0002.jpeg,
                                         ...,
                                         v_BabyCrawling_g09_c02_0248.jpeg]
 vector-frme-values_v_BabyCrawling_g09_c02 = [[0.77,...,0.28], ..., [0.18,...,0.17]]

我们查找每个剪辑引用名称并查找其对应的图像:clips_reference_name+'_0001'.png、clips_reference_name+'_0002'.png ... 并将它们附加到同一个向量。

所以，最后我得到 200 个向量，每个向量代表剪辑引用名称的图像名称。

我没有做到什么？

我在创建 2*200 个向量时陷入困境(然后每个向量的项目数根据描述剪辑引用名称的图像数量而变化)。

如何使用剪辑引用名称来命名每个矢量。向量采用整数索引而不是字符串。

我发现做一个字典，其中键代表剪辑引用名称，每个剪辑引用名称的值是与每个剪辑引用名称关联的图像集。因此，对于每个键，我们都有多个值(一组标签和一组帧值(每个标签为 2048 的一维向量))，这变得难以操作。

Answer 1

如果我没理解错的话，那么您将尝试按引用名称对标签和框架进行分组，引用名称对应于标签名称的第一部分(直到倒数第二个 _)。

然后您可以为这两个组创建字典。

grouped_labels = {}
grouped_frames = {}

然后用“组 key ”填充，如上所述。

for label, frames in dataset.items():
    key = label.rsplit('_', 1)[0]
    grouped_labels.setdefault(key, []).append(label)
    grouped_frames.setdefault(key, []).append(frames)

最后，您可以通过访问轻松获取组的标签和框架:

for crn in clips_reference_name:
    crn_labels = grouped_labels.get(crn, [])
    crn_frames = grouped_frames.get(crn, [])
    # do something with group's labels and frames...