mindvideo.data¶

Acvivitynet¶

class mindvideo.data.Activitynet(path, split=”train”, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=True, batch_size=16, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”, “discrete”, or “align” fetch. Default: “align”.
align(boolean): The video contains multiple actions.Default: True.
batch_size (int): Batch size of dataset. Default:16.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.

Returns:

None

def mindvideo.data.Activitynet.index2label()

Get the mapping of indexes and labels.

Returns:

The mapping of indexes and labels

ParseActivitynet¶

class ParseActivitynet()

Parse Activitynet dataset.

-Base: ParseDataset

def mindvideo.data.ParseActivitynet.loadjson()

Parse json file.

Returns:

cls2index: dictionary
index2cls: list

def mindvideo.data.ParseActivitynet.parse_dataset()

Traverse the Activitynet dataset file to get the path and label.

Returns:

video_path: list
video_label: list

Charades¶

class mindvideo.data.Charades(path, split=”train”, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=True, batch_size=16, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”, “discrete”, or “align” fetch. Default: “align”.
align(boolean): The video contains multiple actions.Default: True.
batch_size (int): Batch size of dataset. Default:32.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.

Returns:

None

def mindvideo.data.Charades.index2label()

Get the mapping of indexes and labels.

Returns:

The mapping of indexes and labels

ParseCharades¶

class ParseCharades()

Parse Charades dataset.

-Base: ParseDataset

def mindvideo.data.ParseCharades.parse_dataset()

Traverse the Charades dataset file to get the path and label.

Returns:

video_path: list
video_label: list

CollectiveActivity¶

class mindvideo.data.CollectiveActivity(path, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=False, batch_size=1, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False, suffix=”picture”)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”, “discrete”, or “align” fetch. Default: “align”.
align(boolean): The video contains multiple actions.Default: True.
batch_size (int): Batch size of dataset. Default:32.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.
suffix(str): Storage format of video. Default: “picture”.

Returns:

None

ParseCollectiveActivity¶

class ParseCollectiveActivity()

Parse CollectiveActivity dataset.

-Base: ParseDataset

def mindvideo.data.ParseCollectiveActivity.parse_dataset()

Traverse the CollectiveActivity dataset file to get the path and label.

Returns:

video_path: list
video_label: list

CCV¶

class mindvideo.data.CCV(path, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=False, batch_size=1, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False, suffix=”picture”)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”, “discrete”, or “align” fetch. Default: “align”.
align(boolean): The video contains multiple actions.Default: True.
batch_size (int): Batch size of dataset. Default:1.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.
suffix(str): Storage format of video. Default: “picture”.

Returns:

None

def mindvideo.data.CCV.index2label()

Get the mapping of indexes and labels.

Returns:

The mapping of indexes and labels

ParseCCV¶

class mindvideo.data.ParseCCV()

Parse columbia consumer video dataset.

-Base: ParseDataset

def mindvideo.data.ParseCCV.load_cls_file()

Parse the category file.

Returns:

a list of category name

def mindvideo.data.ParseCCV.parse_dataset()

Traverse the columbia consumer video dataset file to get the path and label.

Returns:

video_path: list
video_label: list

Davis¶

class mindvideo.data.Davis(path, split=”train”, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=True, quality=”480p”, batch_size=16, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”, “discrete”, or “align” fetch. Default: “align”.
align(boolean): The video contains multiple actions.Default: True.
quality(str):The Picture quality,”1080p” or “480p”.Default:”480p”.
batch_size (int): Batch size of dataset. Default:1.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.

Returns:

None

def mindvideo.data.Davis.index2label()

Get the mapping of indexes and labels.

Returns:

The mapping of indexes and labels

ParseDavis¶

class mindvideo.data.ParseDavis()

Parse Davis dataset.

-Base: ParseDataset

def mindvideo.data.ParseDavis.parse_dataset()

Traverse the Davis dataset file to get the path and label.

Returns:

video_path: list
video_label: list

DatasetGenerator¶

class mindvideo.data.DatasetGenerator(path, label, seq=16, mode=”part”, suffix=”video”, align=False, frame_interval=1, num_clips=1)

Dataset generator for getting video path and its corresponding label.

Parameters:

path(list): Video file path list.
label(list): The label of each video,
seq(int): The number of frames of the intercepted video.
mode(str): Frame fetching method, options:[”part”, “discrete”, “average”, “interval”].
suffix(str): Format of video file. options:[”picture”, “video”].
align(boolean): The video contains multiple actions.
frame_interval(int): Interval between sampling frames.
num_clips(int): The number of samples of a video.

Returns:

None

MixJDE¶

class mindvideo.data.MixJDE(data_json, split=”train”, batch_size=1, repeat_num=1, transform=None, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None)

Multi-dataset based on jde datasets.

Parameters:

data_json (str): Path to a json file that have the path to files that have the path to video frames.
split (str): The dataset split supports “train”, or “test”. Default: “train”.
batch_size (int): Batch size of dataset. Default:1.
repeat_num (int): The repeat num of dataset. Default:1.
transform (callable, optional): A function transform that takes in a video. Default:None.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.

Returns:

None

def mindvideo.data.MixJDE.run()

Dataset pipeline.

JDE¶

class mindvideo.data.JDE(seq_path, data_root, split=”train”, transform=None, batch_size=1, repeat_num=1, resize=224, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, schema_json=None, trans_record=None)

Base: VideoDataset

Parameters:

seq_path (str): Path to a file that have the path to video frames.
data_root (str): Path to
split (str): The dataset split supports “train”, or “test”. Default: “train”.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq (int): The number of frames of captured video. Default: 16.
seq_mode (str): The way of capture video frames,”part”, “discrete”, or “align” fetch. Default: “align”.
align (boolean): The video contains multiple actions.Default: False.
batch_size (int): Batch size of dataset. Default:1.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.

Returns:

None

class mindvideo.data.JDE.read_dataset(*args)

Returns:

the path and the label of image: str

ParseJDE¶

class mindvideo.data.ParseJDE()

Parse JDE dataset.

Base: ParseDataset

def mindvideo.data.ParseJDE.parse_dataset()

Traverse the JDE dataset file to get the path and label.

Returns:

video_path: list
video_label: list

Kinetic400¶

class mindvideo.data.Kinetic400(path=path, split=split, load_data=load_data, transform=transform, target_transform=target_transform, seq=seq, seq_mode=seq_mode, align=align, batch_size=batch_size, repeat_num=repeat_num, shuffle=shuffle, num_parallel_workers=num_parallel_workers, num_shards=num_shards, shard_id=shard_id, download=download, frame_interval=frame_interval, num_clips=num_clips)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part” or “discrete” fetch. Default: “part”.
align(boolean): The video contains multiple actions. Default: False.
batch_size (int): Batch size of dataset. Default:32.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel. Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool): Whether to download the dataset. Default: False.
frame_interval (int): Frame interval of the sample strategy. Default: 1.
num_clips (int): Number of clips sampled in one video. Default: 1.

Returns:

None

def mindvideo.data.Kinetic400.index2label()

Get the mapping of indexes and labels.

Returns:

The mapping of indexes and labels

def mindvideo.data.Kinetic400.default_transform()

Set the default transform for Kinetics400 dataset.

Returns:

transform: list

ParseKinetic400¶

class mindvideo.data.ParseKinetic400()

Parse Kinetics400 dataset.

Base: ParseDataset

def mindvideo.data.ParseKinetics400.load_cls_file()

Parse the category file.

Returns:

id2cls: list cls2id: dict

def mindvideo.data.ParseKinetic400.parse_dataset()

Traverse the Kinetics400 dataset file to get the path and label.

Returns:

video_path: list
video_label: list

Kinetic600¶

class mindvideo.data.Kinetic600(path=path, split=split, load_data=load_data, transform=transform, target_transform=target_transform, seq=seq, seq_mode=seq_mode, align=align, batch_size=batch_size, repeat_num=repeat_num, shuffle=shuffle, num_parallel_workers=num_parallel_workers, num_shards=num_shards, shard_id=shard_id, download=download)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” , “val” or “infer”. Default: train.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”) or “discrete” fetch. Default: “part”.
align(boolean): The video contains multiple actions. Default: False.
batch_size (int): Batch size of dataset. Default:32.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.

Returns:

None

def mindvideo.data.Kinetic600.index2label()

Get the mapping of indexes and labels.

Returns:

The mapping of indexes and labels

def mindvideo.data.Kinetic600.default_transform()

Set the default transform for Kinetics400 dataset.

Returns:

transform: list

ParseKinetic600¶

class mindvideo.data.ParseKinetic600()

Parse Kinetics600 dataset.

Base: ParseDataset

def mindvideo.data.ParseKinetics600.load_cls_file()

Parse the category file.

Returns:

id2cls: list cls2id: dict

def mindvideo.data.ParseKinetic600.parse_dataset()

Traverse the Kinetics600 dataset file to get the path and label.

Returns:

video_path: list
video_label: list

DatasetToMR¶

class mindvideo.data.DatasetToMR(load_data, destination, split, partition_number, schema_json, shard_id)

Transform dataset to MindRecord.

def mindvideo.data.DatasetToMR.trans_to_mr()

Execute transformation from dataset to MindRecord.

Return:

filename: str

Dataset¶

class mindvideo.data.Dataset(path: str, split: str, load_data: Callable, batch_size: int, repeat_num: int, shuffle: bool, num_parallel_workers: Optional[int], num_shards: int, shard_id: int, resize: Optional[int] = None, transform: Optional[Callable] = None, target_transform: Optional[Callable] = None, mode: Optional[str] = None, columns_list: Optional[list] = None, schema_json: Optional[dict] = None, trans_record: Optional[bool] = None)

Dataset is the base class for making dataset which are compatible with MindSpore Vision.

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
load_data(callable): The corresponding video data.
batch_size (int): Batch size of dataset. Default:16.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
mode(str): The way of capture video frames,”part”, “discrete”, or “align” fetch. Default: “align”.
columns_list(str): Column names of dataset.
schema_json(dict): Mapping of category and id.
trans_record(bool): Have the transform record or not. Defalt: None.

Return:

None

def mindvideo.data.Dataset.index2label()

Get the mapping of indexes and labels.

def mindvideo.data.Dataset.default_transform()

Default data augmentation.

def mindvideo.data.Dataset.pipelines()

Data augmentation.

def mindvideo.data.Dataset.run()

Dataset pipeline.

ParseDataset¶

class mindvideo.data.ParseDataset(path: str, shard_id: Optional[int] = None)

Parse dataset.

def mindvideo.data.ParseDatasetparse_dataset(*args)

parse dataset from internet or compression file.

Mot16¶

class mindvideo.data.Mot16(path, split=”train”, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=False, batch_size=1, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False, suffix=”picture”)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, or “test”. Default: “train”.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”), “discrete”, or “align” fetch. Default: “align”.
align(boolean): The video contains multiple actions.Default: False.
batch_size (int): Batch size of dataset. Default:1.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.
suffix(str): Storage format of video. Default: “picture”.

Returns:

None

def mindvideo.data.Mot16.index2label()

Get the mapping of indexes and labels.

def mindvideo.data.Mot16.download_dataset()

Download the Mot16 data if it doesn’t exist already.

ParseMot16¶

class mindvideo.data.ParseMot16()

Parse Mot16 dataset.

Base: ParseDataset

def mindvideo.data.ParseMot16.parse_dataset()

Traverse the Mot16 dataset file to get the path and label.

Returns:

video_path: list
video_label: list

MSVD¶

class mindvideo.data.MSVD(path, split=””, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=True, batch_size=16, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”), “discrete”, or “align” fetch. Default: “align”.
align(boolean): The video contains multiple actions.Default: True.
batch_size (int): Batch size of dataset. Default:32.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.

Returns:

None

def mindvideo.data.MSVD.index2label()

Get the mapping of indexes and labels.

def mindvideo.data.MSVD.download_dataset()

Download the MSVD data if it doesn’t exist already.

ParseMSVD¶

class mindvideo.data.ParseMSVD()

Parse MSVD dataset.

Base: ParseDataset

def mindvideo.data.ParseMSVD.load_label()

Parse annotation file.

Returns:

label2index: dict index2label: list

def mindvideo.data.ParseMSVD.parse_dataset()

Traverse the MSVD dataset file to get the path and label.

Returns:

video_path: list
video_label: list

TaskGenerator¶

class mindvideo.data.TaskGenerator(path, cls, n, k, q)

N-way K-Shot Tasks generator for getting video path and its corresponding label. There are N categories in each task, including K labeled samples in each category.

Parameters:

path(str): video file path.
cls(list): the ending index of the video for each category, the index is start from 1.
n(int): the number of categories per task.
k(int): the number of label samples in each category
q(int): the number of unlabeled samples in each category.

Thumos¶

class mindvideo.data.Thumos(path, split=”train”, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=False, batch_size=16, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test”, “val” or “background”. Default: “infer”.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”) or “discrete” fetch. Default: “part”.
align(boolean): The video contains multiple actions. Default: False.
batch_size (int): Batch size of dataset. Default:32.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.

Returns:

None

def mindvideo.data.Thumos.index2label()

Get the mapping of indexes and labels.

ParseThumos¶

class mindvideo.data.ParseThumos()

Parse Thumos dataset.

Base: ParseDataset

def mindvideo.data.ParseThumos.parse_dataset()

Traverse the Thumos dataset file to get the path and label.

Returns:

video_path: list
video_label: list

UbiFights¶

class mindvideo.data.UbiFights(path, split=”test”, transform=None, target_transform=None, seq=16, seq_mode=”part”, align=True, batch_size=16, repeat_num=1, shuffle=None, num_parallel_workers=1, num_shards=None, shard_id=None, download=False)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: None.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”, “discrete”, or “align” fetch. Default: “align”.
align(boolean): The video contains multiple actions.Default: True.
batch_size (int): Batch size of dataset. Default:32.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.

Returns:

None

def mindvideo.data.UbiFights.index2label()

Get the mapping of indexes and labels.

def mindvideo.data.UbiFights.download_dataset()

Download the UBI-fights data if it doesn’t exist already.

ParseUbiFights¶

class mindvideo.data.ParseUbiFights()

Parse UbiFights dataset.

Base: ParseDataset

def mindvideo.data.ParseUbiFights.parse_dataset()

Traverse the UbiFights dataset file to get the path and label.

Returns:

video_path: list
video_label: list

UCF101¶

class mindvideo.data.UCF101(path: str, split: str = “train”, transform: Optional[Callable] = None, target_transform: Optional[Callable] = None, seq: int = 16, seq_mode: str = “part”, align: bool = False, batch_size: int = 16, repeat_num: int = 1, shuffle: Optional[bool] = None, num_parallel_workers: int = 1, num_shards: Optional[bool] = None, shard_id: Optional[bool] = None, download: bool = False, suffix: str = “video”, task_num: int = 0, task_n: int = 0, task_k: int = 0, task_q: int = 0)

Base: VideoDataset

Parameters:

path (string): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: “train”.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part”) or “discrete” fetch. Default: “part”.
align(boolean): The video contains multiple actions. Default: False.
batch_size (int): Batch size of dataset. Default:16.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel.Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.
suffix(str): Video format to be processed. Optional:(”video”, “picture”, “task”). Default:”video”.
task_num(int): Number of tasks in few shot learning. Default: 0.
task_n(int): Number of categories per task in few shot learning. Default:0.
task_k(int): Number of support sets per task in few shot learning. Default:0.
task_q(int): Number of query sets per task in few shot learning. Default:0.

Returns:

None

def mindvideo.data.UCF101.index2label()

Get the mapping of indexes and labels.

def mindvideo.data.UCF101.download_dataset()

Download the UCF101 data if it doesn’t exist already.

def mindvideo.data.UCF101.default_transform()

Set the default transform for UCF101 dataset.

def mindvideo.data.UCF101.create_task()

Create task list in few shot learning.

ParseUCF101¶

class mindvideo.data.ParseUCF101()

Parse UCF101 dataset.

Base: ParseDataset

def mindvideo.data.ParseUCF101.parse_dataset()

Traverse the UCF101 dataset file to get the path and label.

Returns:

video_path: list
video_label: list

def mindvideo.data.ParseUCF101.load_cls()

Parse category file.

def mindvideo.data.ParseUCF101.modify_struct()

If there is no category subdirectory in the folder, modify the file structure.

VideoDataset¶

class mindvideo.data.VideoDataset(path: str, split: str, load_data: Union[Callable, Tuple], transform: Optional[Callable], target_transform: Optional[Callable], seq: int, seq_mode: str, align: bool, batch_size: int, repeat_num: int, shuffle: bool, num_parallel_workers: Optional[int], num_shards: int, shard_id: int, download: bool, columns_list: List = [’video’, ‘label’], suffix: str = “video”, frame_interval: int = 1, num_clips: int = 1)

VideoDataset is the base class for making video dataset which are compatible with MindSpore Vision.

Base: Dataset

Parameters:

path (str): Root directory of the Mnist dataset or inference image.
split (str): The dataset split supports “train”, “test” or “infer”. Default: “infer”.
transform (callable, optional): A function transform that takes in a video. Default:None.
target_transform (callable, optional): A function transform that takes in a label. Default: None.
seq(int): The number of frames of captured video. Default: 16.
seq_mode(str): The way of capture video frames,”part” or “discrete” fetch. Default: “part”.
align(bool): The video contains multiple actions. Default: False.
batch_size (int): Batch size of dataset. Default:32.
repeat_num (int): The repeat num of dataset. Default:1.
shuffle (bool, optional): Whether or not to perform shuffle on the dataset. Default:None.
num_parallel_workers (int): Number of subprocess used to fetch the dataset in parallel. Default: 1.
num_shards (int, optional): Number of shards that the dataset will be divided into. Default: None.
shard_id (int, optional): The shard ID within num_shards. Default: None.
download (bool) : Whether to download the dataset. Default: False.
frame_interval(int):The number of frame interval when reading video. Default: 1.
num_clips(int):The number of video clips read per video.

def mindvideo.data.VideoDataset.index2label()

Get the mapping of indexes and labels.

def mindvideo.data.VideoDataset.download_dataset()

Download the VideoDataset data if it doesn’t exist already.

def mindvideo.data.VideoDataset.default_transform()

Set the default transform for VideoDataset dataset.

check_file_exist¶

def mindvideo.data.check_file_exist(file_name: str)

Check the input filename is exist or not.

Parameters:

file_name (str): File name.

Returns:

None

Raises:

FileNotFoundError: If file is not exist, print “File {file_name} does not exist.”

check_file_valid¶

def mindvideo.data.check_file_valid(filename: str, extension: Tuple[str, …])

Check image file is valid through the extension.

Parameters:

filename (str): File name.
extension (Tuple[str, …]): Extension of files.

Returns:

Str

check_dir_exist¶

def mindvideo.data.check_dir_exist(dir_name: str)

Check the input directory is exist or not.

Parameters:

dir_name (str): Name of directory.

Returns:

None

Raises:

FileNotFoundError: If the directory is not exist, print “Directory {dir_name} does not exist.”

save_json_file¶

def mindvideo.data.save_json_file(filename: str, data: Dicts)

Save json file.

Parameters:

filename (str): File to be saved.
data (dict): Data of json file.

Returns:

None

load_json_file¶

def mindvideo.data.load_json_file(filename: str)

Load json file.

Parameters:

filename (str): File to be loaded.

Returns:

None

detect_file_type¶

def mindvideo.data.detect_file_type(filename: str)

Detect file type by suffixes and return tuple(suffix, archive_type, compression).

Parameters:

filename (str): File to be detected.

Returns:

Tuple(suffix, archive_type, compression)