VideoMAE (large-sized model, pre-trained only)

VideoMAE model pre-trained on Kinetics-400 for 1600 epochs in a self-supervised way. It was introduced in the paper VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training by Tong et al. and first released in this repository.

Disclaimer: The team releasing VideoMAE did not write a model card for this model so this model card has been written by the Hugging Face team.

Model description

VideoMAE is an extension of Masked Autoencoders (MAE) to video. The architecture of the model is very similar to that of a standard Vision Transformer (ViT), with a decoder on top for predicting pixel values for masked patches.

Videos are presented to the model as a sequence of fixed-size patches (resolution 16x16), which are linearly embedded. One also adds a [CLS] token to the beginning of a sequence to use it for classification tasks. One also adds fixed sinus/cosinus position embeddings before feeding the sequence to the layers of the Transformer encoder.

By pre-training the model, it learns an inner representation of videos that can then be used to extract features useful for downstream tasks: if you have a dataset of labeled videos for instance, you can train a standard classifier by placing a linear layer on top of the pre-trained encoder. One typically places a linear layer on top of the [CLS] token, as the last hidden state of this token can be seen as a representation of an entire video.

Intended uses & limitations

You can use the raw model for predicting pixel values for masked patches of a video, but it's mostly intended to be fine-tuned on a downstream task. See the model hub to look for fine-tuned versions on a task that interests you.

How to use

Here is how to use this model to predict pixel values for randomly masked patches:

from transformers import VideoMAEImageProcessor, VideoMAEForPreTraining
import numpy as np
import torch

num_frames = 16
video = list(np.random.randn(16, 3, 224, 224))

processor = VideoMAEImageProcessor.from_pretrained("MCG-NJU/videomae-large")
model = VideoMAEForPreTraining.from_pretrained("MCG-NJU/videomae-large")

pixel_values = processor(video, return_tensors="pt").pixel_values

num_patches_per_frame = (model.config.image_size // model.config.patch_size) ** 2
seq_length = (num_frames // model.config.tubelet_size) * num_patches_per_frame
bool_masked_pos = torch.randint(0, 2, (1, seq_length)).bool()

outputs = model(pixel_values, bool_masked_pos=bool_masked_pos)
loss = outputs.loss

For more code examples, we refer to the documentation.

Training data

(to do, feel free to open a PR)

Training procedure

Preprocessing

(to do, feel free to open a PR)

Pretraining

(to do, feel free to open a PR)

Evaluation results

(to do, feel free to open a PR)

BibTeX entry and citation info

misc{https://doi.org/10.48550/arxiv.2203.12602,
  doi = {10.48550/ARXIV.2203.12602},
  url = {https://arxiv.org/abs/2203.12602},
  author = {Tong, Zhan and Song, Yibing and Wang, Jue and Wang, Limin},
  keywords = {Computer Vision and Pattern Recognition (cs.CV), FOS: Computer and information sciences, FOS: Computer and information sciences},
  title = {VideoMAE: Masked Autoencoders are Data-Efficient Learners for Self-Supervised Video Pre-Training},
  publisher = {arXiv},
  year = {2022},
  copyright = {Creative Commons Attribution 4.0 International}
}