Video key-frames extractor

This model implements two types of modules: a video frames encoder and the key-frames module. These models are an implementation of a TorchModel.

Pre-trained models

mozuma.models.keyframes.pretrained.torch_keyframes_resnet_imagenet

KeyFrames selector with PyTorch's ResNet pre-trained on ImageNet

Parameters:

Name	Type	Description	Default
resnet_arch	ResNetArchs	Identifier for the ResNet architecture to load. Must be one of: resnet18 resnet34 resnet50 resnet101 resnet152 resnext50_32x4d resnext101_32x8d wide_resnet50_2 wide_resnet101_2	required
fps	float	The number of frames per seconds to extract from the video. Defaults to 1.	1
device	torch.device	Torch device to initialise the model weights	device(type='cpu')

Returns:

Type	Description
KeyFrameSelector	Keyframes model with ResNet pre-trained on ImageNet encoder

mozuma.models.keyframes.pretrained.torch_keyframes_densenet_imagenet

KeyFrames selector with PyTorch DenseNet model pretrained on ImageNet

Parameters:

Name	Type	Description	Default
densenet_arch	DenseNetArch	Identifier for the DenseNet architecture. Must be one of: - densenet121 - densenet161 - densenet169 - densenet201	required
fps	float	The number of frames per seconds to extract from the video. Defaults to 1.	1
device	torch.device	Torch device to initialise the model weights	device(type='cpu')

Returns:

Type	Description
KeyFrameSelector	Keyframes model with DenseNet pre-trained on ImageNet encoder

mozuma.models.keyframes.pretrained.torch_keyframes_densenet_places365

KeyFrames selector with PyTorch DenseNet model pretrained on Places365.

See places365 documentation for more info.

Parameters:

Name	Type	Description	Default
device	torch.device	Torch device to initialise the model weights	device(type='cpu')

Returns:

Type	Description
KeyFrameSelector	Keyframes model with DenseNet model pretrained on Places365

Base key-frames selector model

These models allow to extract key-frames from a video.

mozuma.models.keyframes.selectors.KeyFrameSelector

Video key-frames selector

Attributes:

Name	Type	Description
image_encoder	TorchModel[torch.Tensor, torch.Tensor]	The PyTorch module to encode frames.
fps	float	The number of frames per seconds to extract from the video. Defaults to 1.
device	torch.device	The PyTorch device to initialise the model weights. Defaults to torch.device("cpu").

Base video frames encoder model

mozuma.models.keyframes.encoders.VideoFramesEncoder

Video frames encoder

This module will extract and encode frames of a video using an image_encoder.

Attributes:

Name	Type	Description
image_encoder	TorchModel[torch.Tensor]	The PyTorch module to encode frames
fps	float	The number of frames per seconds to extract from the video. Defaults to 1.
device	torch.device	The PyTorch device to initialise the model weights. Defaults to torch.device("cpu").