TimeSformer
Overview
The TimeSformer model was proposed in TimeSformer: Is Space-Time Attention All You Need for Video Understanding? by Facebook Research. This work is a milestone in action-recognition field being the first video transformer. It inspired many transformer based video understanding and classification papers.
The abstract from the paper is the following:
We present a convolution-free approach to video classification built exclusively on self-attention over space and time. Our method, named “TimeSformer,” adapts the standard Transformer architecture to video by enabling spatiotemporal feature learning directly from a sequence of frame-level patches. Our experimental study compares different self-attention schemes and suggests that “divided attention,” where temporal attention and spatial attention are separately applied within each block, leads to the best video classification accuracy among the design choices considered. Despite the radically new design, TimeSformer achieves state-of-the-art results on several action recognition benchmarks, including the best reported accuracy on Kinetics-400 and Kinetics-600. Finally, compared to 3D convolutional networks, our model is faster to train, it can achieve dramatically higher test efficiency (at a small drop in accuracy), and it can also be applied to much longer video clips (over one minute long). Code and models are available at: this https URL.
Tips:
There are many pretrained variants. Select your pretrained model based on the dataset it is trained on. Moreover, the number of input frames per clip changes based on the model size so you should consider this parameter while selecting your pretrained model.
This model was contributed by fcakyon. The original code can be found here.
Documentation resources
TimesformerConfig
class transformers.TimesformerConfig
< source >( image_size = 224 patch_size = 16 num_channels = 3 num_frames = 8 hidden_size = 768 num_hidden_layers = 12 num_attention_heads = 12 intermediate_size = 3072 hidden_act = 'gelu' hidden_dropout_prob = 0.0 attention_probs_dropout_prob = 0.0 initializer_range = 0.02 layer_norm_eps = 1e-06 qkv_bias = True attention_type = 'divided_space_time' drop_path_rate = 0 **kwargs )
Parameters
-
image_size (
int, optional, defaults to 224) — The size (resolution) of each image. -
patch_size (
int, optional, defaults to 16) — The size (resolution) of each patch. -
num_channels (
int, optional, defaults to 3) — The number of input channels. -
num_frames (
int, optional, defaults to 8) — The number of frames in each video. -
hidden_size (
int, optional, defaults to 768) — Dimensionality of the encoder layers and the pooler layer. -
num_hidden_layers (
int, optional, defaults to 12) — Number of hidden layers in the Transformer encoder. -
num_attention_heads (
int, optional, defaults to 12) — Number of attention heads for each attention layer in the Transformer encoder. -
intermediate_size (
int, optional, defaults to 3072) — Dimensionality of the “intermediate” (i.e., feed-forward) layer in the Transformer encoder. -
hidden_act (
strorfunction, optional, defaults to"gelu") — The non-linear activation function (function or string) in the encoder and pooler. If string,"gelu","relu","selu"and"gelu_new"are supported. -
hidden_dropout_prob (
float, optional, defaults to 0.0) — The dropout probabilitiy for all fully connected layers in the embeddings, encoder, and pooler. -
attention_probs_dropout_prob (
float, optional, defaults to 0.0) — The dropout ratio for the attention probabilities. -
initializer_range (
float, optional, defaults to 0.02) — The standard deviation of the truncated_normal_initializer for initializing all weight matrices. -
layer_norm_eps (
float, optional, defaults to 1e-6) — The epsilon used by the layer normalization layers. -
qkv_bias (
bool, optional, defaults toTrue) — Whether to add a bias to the queries, keys and values. -
attention_type (
str, optional, defaults to"divided_space_time") — The attention type to use. Must be one of"divided_space_time","space_only","joint_space_time". -
drop_path_rate (
float, optional, defaults to 0) — The dropout ratio for stochastic depth.
This is the configuration class to store the configuration of a TimesformerModel. It is used to instantiate a TimeSformer model according to the specified arguments, defining the model architecture. Instantiating a configuration with the defaults will yield a similar configuration to that of the TimeSformer facebook/timesformer-base-finetuned-k600 architecture.
Configuration objects inherit from PretrainedConfig and can be used to control the model outputs. Read the documentation from PretrainedConfig for more information.
Example:
>>> from transformers import TimesformerConfig, TimesformerModel
>>> # Initializing a TimeSformer timesformer-base style configuration
>>> configuration = TimesformerConfig()
>>> # Initializing a model from the configuration
>>> model = TimesformerModel(configuration)
>>> # Accessing the model configuration
>>> configuration = model.configTimesformerModel
class transformers.TimesformerModel
< source >( config )
Parameters
- config (TimesformerConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
The bare TimeSformer Model transformer outputting raw hidden-states without any specific head on top. This model is a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >(
pixel_values
output_attentions = None
output_hidden_states = None
return_dict = None
)
→
transformers.modeling_outputs.BaseModelOutput or tuple(torch.FloatTensor)
Parameters
-
pixel_values (
torch.FloatTensorof shape(batch_size, num_frames, num_channels, height, width)) — Pixel values. Pixel values can be obtained using AutoImageProcessor. See VideoMAEImageProcessor.preprocess() for details. -
output_attentions (
bool, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentionsunder returned tensors for more detail. -
output_hidden_states (
bool, optional) — Whether or not to return the hidden states of all layers. Seehidden_statesunder returned tensors for more detail. -
return_dict (
bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple.
Returns
transformers.modeling_outputs.BaseModelOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.BaseModelOutput or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (TimesformerConfig) and inputs.
-
last_hidden_state (
torch.FloatTensorof shape(batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model. -
hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each layer) of shape(batch_size, sequence_length, hidden_size).Hidden-states of the model at the output of each layer plus the optional initial embedding outputs.
-
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, sequence_length, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
The TimesformerModel forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Examples:
>>> import av
>>> import numpy as np
>>> from transformers import AutoImageProcessor, TimesformerModel
>>> from huggingface_hub import hf_hub_download
>>> np.random.seed(0)
>>> def read_video_pyav(container, indices):
... '''
... Decode the video with PyAV decoder.
... Args:
... container (`av.container.input.InputContainer`): PyAV container.
... indices (`List[int]`): List of frame indices to decode.
... Returns:
... result (np.ndarray): np array of decoded frames of shape (num_frames, height, width, 3).
... '''
... frames = []
... container.seek(0)
... start_index = indices[0]
... end_index = indices[-1]
... for i, frame in enumerate(container.decode(video=0)):
... if i > end_index:
... break
... if i >= start_index and i in indices:
... frames.append(frame)
... return np.stack([x.to_ndarray(format="rgb24") for x in frames])
>>> def sample_frame_indices(clip_len, frame_sample_rate, seg_len):
... converted_len = int(clip_len * frame_sample_rate)
... end_idx = np.random.randint(converted_len, seg_len)
... start_idx = end_idx - converted_len
... indices = np.linspace(start_idx, end_idx, num=clip_len)
... indices = np.clip(indices, start_idx, end_idx - 1).astype(np.int64)
... return indices
>>> # video clip consists of 300 frames (10 seconds at 30 FPS)
>>> file_path = hf_hub_download(
... repo_id="nielsr/video-demo", filename="eating_spaghetti.mp4", repo_type="dataset"
... )
>>> container = av.open(file_path)
>>> # sample 8 frames
>>> indices = sample_frame_indices(clip_len=8, frame_sample_rate=4, seg_len=container.streams.video[0].frames)
>>> video = read_video_pyav(container, indices)
>>> image_processor = AutoImageProcessor.from_pretrained("MCG-NJU/videomae-base")
>>> model = TimesformerModel.from_pretrained("facebook/timesformer-base-finetuned-k400")
>>> # prepare video for the model
>>> inputs = image_processor(list(video), return_tensors="pt")
>>> # forward pass
>>> outputs = model(**inputs)
>>> last_hidden_states = outputs.last_hidden_state
>>> list(last_hidden_states.shape)
[1, 1569, 768]TimesformerForVideoClassification
class transformers.TimesformerForVideoClassification
< source >( config )
Parameters
- config (TimesformerConfig) — Model configuration class with all the parameters of the model. Initializing with a config file does not load the weights associated with the model, only the configuration. Check out the from_pretrained() method to load the model weights.
TimeSformer Model transformer with a video classification head on top (a linear layer on top of the final hidden state of the [CLS] token) e.g. for ImageNet. This model is a PyTorch torch.nn.Module subclass. Use it as a regular PyTorch Module and refer to the PyTorch documentation for all matter related to general usage and behavior.
forward
< source >(
pixel_values: typing.Optional[torch.Tensor] = None
labels: typing.Optional[torch.Tensor] = None
output_attentions: typing.Optional[bool] = None
output_hidden_states: typing.Optional[bool] = None
return_dict: typing.Optional[bool] = None
)
→
transformers.modeling_outputs.ImageClassifierOutput or tuple(torch.FloatTensor)
Parameters
-
pixel_values (
torch.FloatTensorof shape(batch_size, num_frames, num_channels, height, width)) — Pixel values. Pixel values can be obtained using AutoImageProcessor. See VideoMAEImageProcessor.preprocess() for details. -
output_attentions (
bool, optional) — Whether or not to return the attentions tensors of all attention layers. Seeattentionsunder returned tensors for more detail. -
output_hidden_states (
bool, optional) — Whether or not to return the hidden states of all layers. Seehidden_statesunder returned tensors for more detail. -
return_dict (
bool, optional) — Whether or not to return a ModelOutput instead of a plain tuple. -
labels (
torch.LongTensorof shape(batch_size,), optional) — Labels for computing the image classification/regression loss. Indices should be in[0, ..., config.num_labels - 1]. Ifconfig.num_labels == 1a regression loss is computed (Mean-Square loss), Ifconfig.num_labels > 1a classification loss is computed (Cross-Entropy).
Returns
transformers.modeling_outputs.ImageClassifierOutput or tuple(torch.FloatTensor)
A transformers.modeling_outputs.ImageClassifierOutput or a tuple of
torch.FloatTensor (if return_dict=False is passed or when config.return_dict=False) comprising various
elements depending on the configuration (TimesformerConfig) and inputs.
-
loss (
torch.FloatTensorof shape(1,), optional, returned whenlabelsis provided) — Classification (or regression if config.num_labels==1) loss. -
logits (
torch.FloatTensorof shape(batch_size, config.num_labels)) — Classification (or regression if config.num_labels==1) scores (before SoftMax). -
hidden_states (
tuple(torch.FloatTensor), optional, returned whenoutput_hidden_states=Trueis passed or whenconfig.output_hidden_states=True) — Tuple oftorch.FloatTensor(one for the output of the embeddings, if the model has an embedding layer, + one for the output of each stage) of shape(batch_size, sequence_length, hidden_size). Hidden-states (also called feature maps) of the model at the output of each stage. -
attentions (
tuple(torch.FloatTensor), optional, returned whenoutput_attentions=Trueis passed or whenconfig.output_attentions=True) — Tuple oftorch.FloatTensor(one for each layer) of shape(batch_size, num_heads, patch_size, sequence_length).Attentions weights after the attention softmax, used to compute the weighted average in the self-attention heads.
The TimesformerForVideoClassification forward method, overrides the __call__ special method.
Although the recipe for forward pass needs to be defined within this function, one should call the Module
instance afterwards instead of this since the former takes care of running the pre and post processing steps while
the latter silently ignores them.
Examples:
>>> import av
>>> import torch
>>> import numpy as np
>>> from transformers import AutoImageProcessor, TimesformerForVideoClassification
>>> from huggingface_hub import hf_hub_download
>>> np.random.seed(0)
>>> def read_video_pyav(container, indices):
... '''
... Decode the video with PyAV decoder.
... Args:
... container (`av.container.input.InputContainer`): PyAV container.
... indices (`List[int]`): List of frame indices to decode.
... Returns:
... result (np.ndarray): np array of decoded frames of shape (num_frames, height, width, 3).
... '''
... frames = []
... container.seek(0)
... start_index = indices[0]
... end_index = indices[-1]
... for i, frame in enumerate(container.decode(video=0)):
... if i > end_index:
... break
... if i >= start_index and i in indices:
... frames.append(frame)
... return np.stack([x.to_ndarray(format="rgb24") for x in frames])
>>> def sample_frame_indices(clip_len, frame_sample_rate, seg_len):
... converted_len = int(clip_len * frame_sample_rate)
... end_idx = np.random.randint(converted_len, seg_len)
... start_idx = end_idx - converted_len
... indices = np.linspace(start_idx, end_idx, num=clip_len)
... indices = np.clip(indices, start_idx, end_idx - 1).astype(np.int64)
... return indices
>>> # video clip consists of 300 frames (10 seconds at 30 FPS)
>>> file_path = hf_hub_download(
... repo_id="nielsr/video-demo", filename="eating_spaghetti.mp4", repo_type="dataset"
... )
>>> container = av.open(file_path)
>>> # sample 8 frames
>>> indices = sample_frame_indices(clip_len=8, frame_sample_rate=1, seg_len=container.streams.video[0].frames)
>>> video = read_video_pyav(container, indices)
>>> image_processor = AutoImageProcessor.from_pretrained("MCG-NJU/videomae-base-finetuned-kinetics")
>>> model = TimesformerForVideoClassification.from_pretrained("facebook/timesformer-base-finetuned-k400")
>>> inputs = image_processor(list(video), return_tensors="pt")
>>> with torch.no_grad():
... outputs = model(**inputs)
... logits = outputs.logits
>>> # model predicts one of the 400 Kinetics-400 classes
>>> predicted_label = logits.argmax(-1).item()
>>> print(model.config.id2label[predicted_label])
eating spaghetti