Abstract: Video anomaly detection (VAD) aims to temporally locate abnormal events in a video. Existing works mostly rely on training deep models to learn the distribution of normality with either video-level supervision, one-class supervision, or in an unsupervised setting. Training-based methods are prone to be domain-specific, thus being costly for practical deployment as any domain change will involve data collection and model training. In this paper, we radically depart from previous efforts and propose LAnguage-based VAD (LAVAD), a method tackling VAD in a novel, training-free paradigm, exploiting the capabilities of pre-trained large language models (LLMs) and existing vision-language models (VLMs). We leverage VLM-based captioning models to generate textual descriptions for each frame of any test video. With the textual scene description, we then devise a prompting mechanism to unlock the capability of LLMs in terms of temporal aggregation and anomaly score estimation, turning LLMs into an effective video anomaly detector. We further leverage modality-aligned VLMs and propose effective techniques based on cross-modal similarity for cleaning noisy captions and refining the LLM-based anomaly scores. We evaluate LAVAD on two large datasets featuring real-world surveillance scenarios (UCF-Crime and XD-Violence), showing that it outperforms both unsupervised and one-class methods without requiring any training or data collection.

Setup

We recommend the use of a Linux machine with CUDA compatible GPUs. We used 2x NVIDIA A100 GPUs with 64GB. We provide both a Conda environment and a Dockerfile to configure the required libraries.

Clone the repo with:

git clone https://github.com/lucazanella/lavad.git
cd lavad

Conda

The environment can be installed and activated with:

conda create --name lavad python=3.10
conda activate lavad
pip install -r requirements.txt

Docker

Alternatively, you can use the Dockerfile to build a docker image with the following command:

docker build -t lavad -f docker/Dockerfile .

Run the Docker container by mounting the project root to /usr/src/app and the folder containing your datasets to /usr/src/datasets with the following command:

docker run --shm-size 64gb --gpus '"device=0,1"' --rm -it -v $(pwd):/usr/src/app -v /path/to/directory/datasets:/usr/src/datasets lavad /bin/bash

Data

Please download the data, including captions, temporal summaries, indexes with their textual embeddings, and scores for the UCF-Crime and XD-Violence datasets, from the links below:

and place them in the /path/to/directory/datasets folder. The code works with pre-extracted frames, which are not provided within the zipped files. You can download the videos from the official websites (UCF-Crime and XD-Violence) and extract the frames yourself using a script similar to scripts/00_extract_frames.sh. Please note that you need to change the paths within the files included in the annotations folder of each dataset accordingly.

Pretrained models

We employ BLIP-2 as the captioning module, Llama 2 as our large language model (LLM) module, and multimodal encoders provided by ImageBind. While the download of BLIP-2 and ImageBind models should start automatically, you need to download Llama 2. Please download llama-2-13b-chat from the official repository by following the download instructions and place the model and the tokenizer in libs/llama.

Evaluation

Are you interested in running LAVAD on UCF-Crime and XD-Violence?

The slurm scripts used to generate captions, temporal summaries, indexes with their textual embeddings, and scores for UCF-Crime can be found in slurm/ucf_crime, while those used for XD-Violence can be found in slurm/xd_violence. The scripts are numbered for ease of execution. Each script requires completion of the previous steps before execution.

where you need to replace <dataset> with either ucf_crime or xd_violence.

Custom Inference

Are you interested in running LAVAD on your videos?

You can save the videos in a videos folder and run the scripts in scripts one at a time by updating the variables of interest (e.g., dataset_dir and fps). Please note that you first need to extract the frames using scripts/00_extract_frames.sh.

Citation

Please consider citing our paper in your publications if the project helps your research.

@article{zanella2024harnessing,
  title={Harnessing Large Language Models for Training-free Video Anomaly Detection},
  author={Zanella, Luca and Menapace, Willi and Mancini, Massimiliano and Wang, Yiming and Ricci, Elisa},
  journal={arXiv preprint arXiv:2404.01014},
  year={2024}
}