IPAD: Industrial Process Anomaly Detection Dataset

TL;DR

IPAD introduces the first industrial-focused video anomaly detection dataset with real and synthetic footage across 16 devices and explicit periodicity annotations. It proposes a reconstruction-based model that integrates periodic information via a periodic memory module and a sliding window inspection, augmented by LoRA-style adapters for efficient real-world deployment. The approach demonstrates improved anomaly detection performance on IPAD and shows favorable synthetic-to-real migration, with ablations confirming the value of periodic information. Collectively, the dataset and method aim to accelerate robust VAD deployment in smart factories by leveraging cyclic industrial processes and efficient fine-tuning.

Abstract

Video anomaly detection (VAD) is a challenging task aiming to recognize anomalies in video frames, and existing large-scale VAD researches primarily focus on road traffic and human activity scenes. In industrial scenes, there are often a variety of unpredictable anomalies, and the VAD method can play a significant role in these scenarios. However, there is a lack of applicable datasets and methods specifically tailored for industrial production scenarios due to concerns regarding privacy and security. To bridge this gap, we propose a new dataset, IPAD, specifically designed for VAD in industrial scenarios. The industrial processes in our dataset are chosen through on-site factory research and discussions with engineers. This dataset covers 16 different industrial devices and contains over 6 hours of both synthetic and real-world video footage. Moreover, we annotate the key feature of the industrial process, ie, periodicity. Based on the proposed dataset, we introduce a period memory module and a sliding window inspection mechanism to effectively investigate the periodic information in a basic reconstruction model. Our framework leverages LoRA adapter to explore the effective migration of pretrained models, which are initially trained using synthetic data, into real-world scenarios. Our proposed dataset and method will fill the gap in the field of industrial video anomaly detection and drive the process of video understanding tasks as well as smart factory deployment.

Figures (4)

• Figure 3: Dataset statistics. The bar chart shows the percentage of train data and test data in different cases. The specific devices in each case are given in the table.
• Figure 4: The proposed reconstruction-based model with periodic information. The model undergoes an initial pre-training phase using synthetic data. Subsequently, the encoder segment is fine-tuned through a combination of real data and LoRA hu2021lora, enhancing its performance and adaptability. The features of the input video clip are first extracted by the video swin transformer to get $F$. Then addressing to the memory module and period classification are performed separately to get the initial weights $W$ and period relative position ${t}_{p}$. Then the weights are updated to get $\hat{w}$, which in turn is computed to get the output feature $\hat{F}$. Finally, the reconstructed video frame is obtained using the I3D decoder.
• Figure 5: Sliding window. The sequence represents an action cycle, where each square represents an input video clip.
• Figure 6: Anomaly score curves. The first row shows the detection results of two anomalies. The second row shows the case of light change and camera jitter, and it can be seen that our method has lower anomaly scores, which is closer to GT.