These are the results obtained from:
1) GLASS:
2) SimpleNet
AnomalyFTP is an open-source implementation focused on unsupervised anomaly detection for industrial visual inspection tasks. The repository provides tools and code for detecting visual anomalies in industrial settings, where abnormal samples are rare and defects can range from subtle scratches to significant structural issues.
- Implements state-of-the-art unsupervised anomaly detection methods tailored for industrial images.
- Designed for industrial inspection scenarios where labeled anomalies are scarce.
- Supports evaluation and benchmarking on standard datasets.
The MVTec Anomaly Detection (MVTec AD) dataset is a widely used benchmark for evaluating unsupervised anomaly detection methods, particularly in the context of industrial visual inspection. It is specifically designed to reflect real-world manufacturing scenarios, where defective samples are scarce, and models are typically trained only on normal (defect-free) images.
The repository explores advanced anomaly detection techniques, including models inspired by recent research GLASS and SimpleNet.
GLASS (Global and Local Anomaly co-Synthesis Strategy) is a unified framework for synthesizing a broader and more controllable range of anomalies at both the feature and image levels. It consists of:
- Global Anomaly Synthesis (GAS): Synthesizes weak, near-in-distribution anomalies at the feature level using Gaussian noise guided by gradient ascent and truncated projection. This approach enhances the detection of subtle defects that are close to normal samples.
- Local Anomaly Synthesis (LAS): Generates strong, far-from-distribution anomalies at the image level by overlaying textures, providing a diverse set of synthetic anomalies.
GLASS achieves state-of-the-art results on major industrial benchmarks such as MVTec AD (detection AUROC of 99.9%), VisA, and MPDD, and it excels in weak defect detection. Its effectiveness and efficiency have been validated in real-world industrial applications.
SimpleNet is a lightweight and application-friendly network for image anomaly detection and localization. It includes:
- A pre-trained feature extractor for local feature extraction.
- A shallow feature adapter for domain adaptation.
- An anomaly feature generator that injects Gaussian noise during training.
- A binary anomaly discriminator for distinguishing normal and anomalous features.
SimpleNet achieves high accuracy and fast inference, making it suitable for deployment in industrial environments.
- Clone the repository:
git clone https://github.com/nabayansaha/AnomalyFTP.git
cd AnomalyFTP-
Install dependencies:
- Install required packages (see requirements.txt).
-
Prepare your data:
- Organize your industrial inspection images as described in the documentation.
-
Run the code:
- Follow the usage instructions in the repository to train and evaluate anomaly detection models.
- GLASS:
| Category | I-AUROC (%) | P-AUROC (%) |
|---|---|---|
| Bottle | 99.60 | 83.76 |
| Cable | 87.95 | 71.55 |
| Capsule | 94.34 | 63.60 |
| Carpet | 93.24 | 93.51 |
| Grid | 97.24 | 88.02 |
| Hazelnut | 94.61 | 73.80 |
| Leather | 99.80 | 97.50 |
| Metal Nut | 99.56 | 70.58 |
| Pill | 91.49 | 62.11 |
| Screw | 86.62 | 68.41 |
| Tile | 100.00 | 87.81 |
| Toothbrush | 87.50 | 90.57 |
| Transistor | 86.50 | 44.69 |
| Wood | 98.51 | 83.10 |
| Zipper | 99.89 | 81.96 |
| Average | 94.91 | 77.89 |
- SimpleNet:
| Category | I-AUROC (%) | P-AUROC (%) |
|---|---|---|
| Bottle | 98.49 | 82.75 |
| Cable | 88.08 | 79.62 |
| Capsule | 90.03 | 95.23 |
| Carpet | 93.86 | 75.40 |
| Grid | 92.94 | 87.87 |
| Hazelnut | 60.04 | 68.99 |
| Leather | 84.14 | 87.19 |
| Metal Nut | 71.36 | 89.96 |
| Pill | 71.09 | 86.19 |
| Screw | 69.34 | 76.93 |
| Tile | 78.86 | 67.79 |
| Toothbrush | 80.83 | 85.73 |
| Transistor | 77.08 | 79.08 |
| Wood | 92.98 | 76.82 |
| Zipper | 98.53 | 92.17 |
| Average | 83.66 | 81.99 |
This project is released under the Apache License
- SimpleNet: A Simple Network for Image Anomaly Detection and Localization
- A Unified Anomaly Synthesis Strategy with Gradient Ascent for Industrial Anomaly Detection and Localization
For more details, usage instructions, and contribution guidelines, please refer to the repository documentation.