The DINOv2 model was proposed in DINOv2: Learning Robust Visual Features without Supervision by Maxime Oquab, Timothée Darcet, Théo Moutakanni, Huy Vo, Marc Szafraniec, Vasil Khalidov, Pierre Fernandez, Daniel Haziza, Francisco Massa, Alaaeldin El-Nouby, Mahmoud Assran, Nicolas Ballas, Wojciech Galuba, Russell Howes, Po-Yao Huang, Shang-Wen Li, Ishan Misra, Michael Rabbat, Vasu Sharma, Gabriel Synnaeve, Hu Xu, Hervé Jegou, Julien Mairal, Patrick Labatut, Armand Joulin, Piotr Bojanowski. DINOv2 is an upgrade of DINO, a self-supervised method applied on Vision Transformers. This method enables all-purpose visual features, i.e., features that work across image distributions and tasks without finetuning.
The abstract from the paper is the following:
The recent breakthroughs in natural language processing for model pretraining on large quantities of data have opened the way for similar foundation models in computer vision. These models could greatly simplify the use of images in any system by producing all-purpose visual features, i.e., features that work across image distributions and tasks without finetuning. This work shows that existing pretraining methods, especially self-supervised methods, can produce such features if trained on enough curated data from diverse sources. We revisit existing approaches and combine different techniques to scale our pretraining in terms of data and model size. Most of the technical contributions aim at accelerating and stabilizing the training at scale. In terms of data, we propose an automatic pipeline to build a dedicated, diverse, and curated image dataset instead of uncurated data, as typically done in the self-supervised literature. In terms of models, we train a ViT model (Dosovitskiy et al., 2020) with 1B parameters and distill it into a series of smaller models that surpass the best available all-purpose features, OpenCLIP (Ilharco et al., 2021) on most of the benchmarks at image and pixel levels.
This model was contributed by nielsr. The original code can be found here.
The model can be traced using torch.jit.trace which leverages JIT compilation to optimize the model making it faster to run. Note this still produces some mis-matched elements and the difference between the original model and the traced model is of the order of 1e-4.
import torch
from transformers import AutoImageProcessor, AutoModel
from PIL import Image
import requests
url = 'http://images.cocodataset.org/val2017/000000039769.jpg'
image = Image.open(requests.get(url, stream=True).raw)
processor = AutoImageProcessor.from_pretrained('facebook/dinov2-base')
model = AutoModel.from_pretrained('facebook/dinov2-base')
inputs = processor(images=image, return_tensors="pt")
outputs = model(**inputs)
last_hidden_states = outputs[0]
# We have to force return_dict=False for tracing
model.config.return_dict = False
with torch.no_grad():
traced_model = torch.jit.trace(model, [inputs.pixel_values])
traced_outputs = traced_model(inputs.pixel_values)
print((last_hidden_states - traced_outputs[0]).abs().max())A list of official Hugging Face and community (indicated by 🌎) resources to help you get started with DPT.
If you’re interested in submitting a resource to be included here, please feel free to open a Pull Request and we’ll review it! The resource should ideally demonstrate something new instead of duplicating an existing resource.
( hidden_size = 768 num_hidden_layers = 12 num_attention_heads = 12 mlp_ratio = 4 hidden_act = 'gelu' hidden_dropout_prob = 0.0 attention_probs_dropout_prob = 0.0 initializer_range = 0.02 layer_norm_eps = 1e-06 image_size = 224 patch_size = 16 num_channels = 3 qkv_bias = True layerscale_value = 1.0 drop_path_rate = 0.0 use_swiglu_ffn = False out_features = None out_indices = None apply_layernorm = True reshape_hidden_states = True **kwargs )
Parameters
int, optional, defaults to 768) —
Dimensionality of the encoder layers and the pooler layer. int, optional, defaults to 12) —
Number of hidden layers in the Transformer encoder. int, optional, defaults to 12) —
Number of attention heads for each attention layer in the Transformer encoder. int, optional, defaults to 4) —
Ratio of the hidden size of the MLPs relative to the hidden_size. str or function, 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. float, optional, defaults to 0.0) —
The dropout probability for all fully connected layers in the embeddings, encoder, and pooler. float, optional, defaults to 0.0) —
The dropout ratio for the attention probabilities. float, optional, defaults to 0.02) —
The standard deviation of the truncated_normal_initializer for initializing all weight matrices. float, optional, defaults to 1e-06) —
The epsilon used by the layer normalization layers. int, optional, defaults to 224) —
The size (resolution) of each image. int, optional, defaults to 16) —
The size (resolution) of each patch. int, optional, defaults to 3) —
The number of input channels. bool, optional, defaults to True) —
Whether to add a bias to the queries, keys and values. float, optional, defaults to 1.0) —
Initial value to use for layer scale. float, optional, defaults to 0.0) —
Stochastic depth rate per sample (when applied in the main path of residual layers). bool, optional, defaults to False) —
Whether to use the SwiGLU feedforward neural network. List[str], optional) —
If used as backbone, list of features to output. Can be any of "stem", "stage1", "stage2", etc.
(depending on how many stages the model has). If unset and out_indices is set, will default to the
corresponding stages. If unset and out_indices is unset, will default to the last stage. Must be in the
same order as defined in the stage_names attribute. List[int], optional) —
If used as backbone, list of indices of features to output. Can be any of 0, 1, 2, etc. (depending on how
many stages the model has). If unset and out_features is set, will default to the corresponding stages.
If unset and out_features is unset, will default to the last stage. Must be in the
same order as defined in the stage_names attribute. bool, optional, defaults to True) —
Whether to apply layer normalization to the feature maps in case the model is used as backbone. bool, optional, defaults to True) —
Whether to reshape the feature maps to 4D tensors of shape (batch_size, hidden_size, height, width) in
case the model is used as backbone. If False, the feature maps will be 3D tensors of shape (batch_size, seq_len, hidden_size). This is the configuration class to store the configuration of a Dinov2Model. It is used to instantiate an Dinov2 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 Dinov2 google/dinov2-base-patch16-224 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 Dinov2Config, Dinov2Model
>>> # Initializing a Dinov2 dinov2-base-patch16-224 style configuration
>>> configuration = Dinov2Config()
>>> # Initializing a model (with random weights) from the dinov2-base-patch16-224 style configuration
>>> model = Dinov2Model(configuration)
>>> # Accessing the model configuration
>>> configuration = model.config( config: Dinov2Config )
Parameters
The bare DINOv2 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.
( pixel_values: Optional = None bool_masked_pos: Optional = None head_mask: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None ) → transformers.modeling_outputs.BaseModelOutputWithPooling or tuple(torch.FloatTensor)
Parameters
torch.FloatTensor of shape (batch_size, num_channels, height, width)) —
Pixel values. Pixel values can be obtained using AutoImageProcessor. See
BitImageProcessor.preprocess() for details. torch.BoolTensor of shape (batch_size, sequence_length)) —
Boolean masked positions. Indicates which patches are masked (1) and which aren’t (0). Only relevant for
pre-training. torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) —
Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]:
bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. Returns
transformers.modeling_outputs.BaseModelOutputWithPooling or tuple(torch.FloatTensor)
A transformers.modeling_outputs.BaseModelOutputWithPooling 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 (Dinov2Config) and inputs.
last_hidden_state (torch.FloatTensor of shape (batch_size, sequence_length, hidden_size)) — Sequence of hidden-states at the output of the last layer of the model.
pooler_output (torch.FloatTensor of shape (batch_size, hidden_size)) — Last layer hidden-state of the first token of the sequence (classification token) after further processing
through the layers used for the auxiliary pretraining task. E.g. for BERT-family of models, this returns
the classification token after processing through a linear layer and a tanh activation function. The linear
layer weights are trained from the next sentence prediction (classification) objective during pretraining.
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.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 Dinov2Model 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.
Example:
>>> from transformers import AutoImageProcessor, Dinov2Model
>>> import torch
>>> from datasets import load_dataset
>>> dataset = load_dataset("huggingface/cats-image")
>>> image = dataset["test"]["image"][0]
>>> image_processor = AutoImageProcessor.from_pretrained("facebook/dinov2-base")
>>> model = Dinov2Model.from_pretrained("facebook/dinov2-base")
>>> inputs = image_processor(image, return_tensors="pt")
>>> with torch.no_grad():
... outputs = model(**inputs)
>>> last_hidden_states = outputs.last_hidden_state
>>> list(last_hidden_states.shape)
[1, 257, 768]( config: Dinov2Config )
Parameters
Dinov2 Model transformer with an image 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.
( pixel_values: Optional = None head_mask: Optional = None labels: Optional = None output_attentions: Optional = None output_hidden_states: Optional = None return_dict: Optional = None ) → transformers.modeling_outputs.ImageClassifierOutput or tuple(torch.FloatTensor)
Parameters
torch.FloatTensor of shape (batch_size, num_channels, height, width)) —
Pixel values. Pixel values can be obtained using AutoImageProcessor. See
BitImageProcessor.preprocess() for details. torch.FloatTensor of shape (num_heads,) or (num_layers, num_heads), optional) —
Mask to nullify selected heads of the self-attention modules. Mask values selected in [0, 1]:
bool, optional) —
Whether or not to return the attentions tensors of all attention layers. See attentions under returned
tensors for more detail. bool, optional) —
Whether or not to return the hidden states of all layers. See hidden_states under returned tensors for
more detail. bool, optional) —
Whether or not to return a ModelOutput instead of a plain tuple. torch.LongTensor of shape (batch_size,), optional) —
Labels for computing the image classification/regression loss. Indices should be in [0, ..., config.num_labels - 1]. If config.num_labels == 1 a regression loss is computed (Mean-Square loss), If
config.num_labels > 1 a 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 (Dinov2Config) and inputs.
loss (torch.FloatTensor of shape (1,), optional, returned when labels is provided) — Classification (or regression if config.num_labels==1) loss.
logits (torch.FloatTensor of shape (batch_size, config.num_labels)) — Classification (or regression if config.num_labels==1) scores (before SoftMax).
hidden_states (tuple(torch.FloatTensor), optional, returned when output_hidden_states=True is passed or when config.output_hidden_states=True) — Tuple of torch.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 when output_attentions=True is passed or when config.output_attentions=True) — Tuple of torch.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 Dinov2ForImageClassification 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.
Example:
>>> from transformers import AutoImageProcessor, Dinov2ForImageClassification
>>> import torch
>>> from datasets import load_dataset
>>> dataset = load_dataset("huggingface/cats-image")
>>> image = dataset["test"]["image"][0]
>>> image_processor = AutoImageProcessor.from_pretrained("facebook/dinov2-small-imagenet1k-1-layer")
>>> model = Dinov2ForImageClassification.from_pretrained("facebook/dinov2-small-imagenet1k-1-layer")
>>> inputs = image_processor(image, return_tensors="pt")
>>> with torch.no_grad():
... logits = model(**inputs).logits
>>> # model predicts one of the 1000 ImageNet classes
>>> predicted_label = logits.argmax(-1).item()
>>> print(model.config.id2label[predicted_label])
tabby, tabby cat