vjepa2 vitl fpc16 256 ssv2

提供商facebook
分类video-classification
许可证mit
下载量325
星标0

简介

V-JEPA 是 Meta 推出的一款基于联合嵌入预测架构的视频表征模型。与传统的生成式模型不同，它通过预测缺失的视频块来学习视频的语义特征，而非重建像素，这使其在处理复杂视频场景时具有极高的效率。该版本采用了 ViT-L 骨干网络并在 SSv2 数据集上进行了预训练，能够精准捕捉视频中的动作演变和时空关系。对于开发者而言，它非常适合作为视频分类、动作识别等下游任务的特征提取器，通过简单的线性探测或微调即可快速上手，是构建高效视频理解应用的有力底座。

核心亮点

  • 基于 V-JEPA 架构，视频语义表征能力极强
  • ViT-L 大模型底座，时空特征提取精度高
  • 适配 SSv2 数据集，擅长复杂动作识别
  • MIT 协议开源，方便商业化集成与二次开发

使用方法

安装依赖
# 安装 Hugging Face transformers
pip install transformers torch
SDK 使用
# 使用 transformers 加载模型
from transformers import AutoModel, AutoTokenizer

model = AutoModel.from_pretrained("facebook/vjepa2-vitl-fpc16-256-ssv2")
tokenizer = AutoTokenizer.from_pretrained("facebook/vjepa2-vitl-fpc16-256-ssv2")

Hugging Face 下载

我们推荐使用命令行或者 Hugging Face Hub SDK 来进行模型的下载。

操作指引：在下载前，请先通过如下命令安装 huggingface_hub：

操作指引 
pip install -U huggingface_hub

命令行下载

下载完整模型库

下载完整模型库 
huggingface-cli download facebook/vjepa2-vitl-fpc16-256-ssv2

下载单个文件到指定本地文件夹（以下载 config.json 到当前路径下 ./dir 目录为例）

下载单个文件到指定本地文件夹（以下载 config.json 到当前路径下 ./dir 目录为例） 
huggingface-cli download facebook/vjepa2-vitl-fpc16-256-ssv2 config.json --local-dir ./dir

更多命令行下载选项，可参见官方文档

SDK 下载

SDK 下载 
# 模型下载
from huggingface_hub import snapshot_download
model_dir = snapshot_download('facebook/vjepa2-vitl-fpc16-256-ssv2')

Git 下载

请确保 lfs 已经被正确安装

Git 下载 
git lfs install
git clone https://huggingface.co/facebook/vjepa2-vitl-fpc16-256-ssv2

如果您希望跳过 lfs 大文件下载，可以使用如下命令

跳过 LFS 
GIT_LFS_SKIP_SMUDGE=1 git clone https://huggingface.co/facebook/vjepa2-vitl-fpc16-256-ssv2

模型文件托管在 Hugging Face Hub，使用 HF CLI / SDK / Git 直接下载，不经过本站。

PyTorch / Transformers 使用

安装 Transformers

安装 Transformers 
pip install -U transformers torch

模型加载和推理

模型加载和推理 
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained('facebook/vjepa2-vitl-fpc16-256-ssv2')
tokenizer = AutoTokenizer.from_pretrained('facebook/vjepa2-vitl-fpc16-256-ssv2')

模型下载

我们推荐使用命令行或者 ModelScope SDK 来进行模型的下载。

操作指引：在下载前，请先通过如下命令安装 ModelScope：

操作指引 
pip install modelscope

命令行下载

下载完整模型库

下载完整模型库 
modelscope download --model facebook/vjepa2-vitl-fpc16-256-ssv2

下载单个文件到指定本地文件夹（以下载 README.md 到当前路径下 dir 目录为例）

下载单个文件到指定本地文件夹（以下载 README.md 到当前路径下 dir 目录为例） 
modelscope download --model facebook/vjepa2-vitl-fpc16-256-ssv2 README.md --local_dir ./dir

更多更丰富的命令行下载选项，可参见具体文档

SDK 下载

SDK 下载 
# 模型下载
from modelscope import snapshot_download
model_dir = snapshot_download('facebook/vjepa2-vitl-fpc16-256-ssv2')

Git 下载

请确保 lfs 已经被正确安装

Git 下载 
git lfs install
git clone https://www.modelscope.cn/facebook/vjepa2-vitl-fpc16-256-ssv2.git

如果您希望跳过 lfs 大文件下载，可以使用如下命令

跳过 LFS 
GIT_LFS_SKIP_SMUDGE=1 git clone https://www.modelscope.cn/facebook/vjepa2-vitl-fpc16-256-ssv2.git

ModelScope 模型页直接下载模型文件；无需将模型文件放在本站服务器。

Notebook 快速开发

下载并安装 ModelScope library

下载并安装 ModelScope library 
pip install "modelscope[audio,cv,nlp,multi-modal,science]" -f https://modelscope.oss-cn-beijing.aliyuncs.com/releases/repo.html

模型加载和推理

模型加载和推理 
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks

p = pipeline('text-generation', 'facebook/vjepa2-vitl-fpc16-256-ssv2')

完整文档

来源: HuggingFace

---
license: mit
pipeline_tag: video-classification
tags:

  • video

library_name: transformers
datasets:
  • HuggingFaceM4/something_something_v2

base_model:
  • facebook/vjepa2-vitl-fpc64-256

---

V-JEPA 2

A frontier video understanding model developed by FAIR, Meta, which extends the pretraining objectives of VJEPA, resulting in state-of-the-art video understanding capabilities, leveraging data and model sizes at scale.
The code is released in this repository.

<div style="background-color: rgba(251, 255, 120, 0.4); padding: 10px; color: black; border-radius: 5px; box-shadow: 0 4px 8px rgba(0,0,0,0.1);">
💡 This is V-JEPA 2 <a href="https://huggingface.co/facebook/vjepa2-vitl-fpc64-256">ViT-L 256</a> model with video classification head pretrained on <a href="https://paperswithcode.com/dataset/something-something-v2" style="color: black;">Something-Something-V2</a> dataset.
</div>
<br></br>

<img src="https://github.com/user-attachments/assets/914942d8-6a1e-409d-86ff-ff856b7346ab">&nbsp;

Installation

To run V-JEPA 2 model, ensure you have installed the latest transformers:

bash
pip install -U git+https://github.com/huggingface/transformers

Video classification code snippet

python
import torch
import numpy as np

from torchcodec.decoders import VideoDecoder
from transformers import AutoVideoProcessor, AutoModelForVideoClassification


device = "cuda" if torch.cuda.is_available() else "cpu"


Load model and video preprocessor

hf_repo = "facebook/vjepa2-vitl-fpc16-256-ssv2"

model = AutoModelForVideoClassification.from_pretrained(hf_repo).to(device)
processor = AutoVideoProcessor.from_pretrained(hf_repo)


To load a video, sample the number of frames according to the model.

video_url = "https://huggingface.co/datasets/nateraw/kinetics-mini/resolve/main/val/bowling/-WH-lxmGJVY_000005_000015.mp4"
vr = VideoDecoder(video_url)
frame_idx = np.arange(0, model.config.frames_per_clip, 8) # you can define more complex sampling strategy
video = vr.get_frames_at(indices=frame_idx).data  # frames x channels x height x width

Preprocess and run inference

inputs = processor(video, return_tensors="pt").to(model.device)
with torch.no_grad():
    outputs = model(**inputs)
logits = outputs.logits

print("Top 5 predicted class names:")
top5_indices = logits.topk(5).indices[0]
top5_probs = torch.softmax(logits, dim=-1).topk(5).values[0]
for idx, prob in zip(top5_indices, top5_probs):
    text_label = model.config.id2label[idx.item()]
    print(f" - {text_label}: {prob:.2f}")

Output:
code
Top 5 predicted class names:
 - Stuffing [something] into [something]: 0.34
 - Putting [something] into [something]: 0.25
 - Putting [something] onto [something]: 0.04
 - Spreading [something] onto [something]: 0.04
 - Closing [something]: 0.03

Citation

code
@techreport{assran2025vjepa2,
  title={V-JEPA~2: Self-Supervised Video Models Enable Understanding, Prediction and Planning},
  author={Assran, Mahmoud and Bardes, Adrien and Fan, David and Garrido, Quentin and Howes, Russell and
  Komeili, Mojtaba and Muckley, Matthew and Rizvi, Ammar and Roberts, Claire and Sinha, Koustuv and Zholus, Artem and
  Arnaud, Sergio and Gejji, Abha and Martin, Ada and Robert Hogan, Francois and Dugas, Daniel and
  Bojanowski, Piotr and Khalidov, Vasil and Labatut, Patrick and Massa, Francisco and Szafraniec, Marc and
  Krishnakumar, Kapil and Li, Yong and Ma, Xiaodong and Chandar, Sarath and Meier, Franziska and LeCun, Yann and
  Rabbat, Michael and Ballas, Nicolas},
  institution={FAIR at Meta},
  year={2025}
}
查看官方来源