Summary of Unified Anomaly Detection Methods on Edge Device Using Knowledge Distillation and Quantization, by Sushovan Jena et al.
Unified Anomaly Detection methods on Edge Device using Knowledge Distillation and Quantization
by Sushovan Jena, Arya Pulkit, Kajal Singh, Anoushka Banerjee, Sharad Joshi, Ananth Ganesh, Dinesh Singh, Arnav Bhavsar
First submitted to arxiv on: 3 Jul 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI); Computational Complexity (cs.CC); Emerging Technologies (cs.ET)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary A unified multi-class approach to visual inspection systems is proposed, aiming to reduce cost and memory requirements for anomaly detection tasks. The study experiments with considering a unified multi-class setup on the MVTec AD dataset, showing that multi-class models perform similarly to one-class models. Lightweight architectures are deployed on CPU and edge devices (NVIDIA Jetson Xavier NX), analyzing quantized models in terms of latency and memory requirements. Quantization-aware training (QAT) and post-training quantization (PTQ) are explored at different precision widths, with QAT compensating for the performance drop in PTQ. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper shows that a unified multi-class approach can be used for anomaly detection tasks. This means that instead of using separate models to detect different types of defects, one model can be used for all types of defects. The study uses a dataset called MVTec AD and shows that this approach works well. It also tests how well the models work on edge devices, which are small computers that are often used in industry settings. |
Keywords
» Artificial intelligence » Anomaly detection » Precision » Quantization
