Updated model checkpoints and descriptions

README.md

@@ -1,3 +1,119 @@
 ---
 license: mit
+language:
+- en
+tags:
+- anomaly-detection
+- computer-vision
+- vae-gan
+- one-class-classification
+- robotics
+- industrial-ai
+- visual-anomaly-detection
+- pytorch
+library_name: pytorch
+pipeline_tag: image-classification
+datasets:
+- hazards-robots-corridor
 ---
+
+# VAE-GAN for Corridor Hazard Anomaly Detection
+
+<img src='AD_Robotics_Hazards/docs/Dataset.svg'>
+
+This repository contains trained **VAE-GAN model checkpoints** for visual anomaly detection in robotic environments. The models were trained using a one-class learning paradigm, where only normal operating conditions are observed during training and anomalous situations are detected during inference through reconstruction-based anomaly scoring.
+
+## Dataset
+
+The models were trained on the **Corridor** scenario from the *Hazards&Robots* dataset.
+
+### Original Dataset
+
+**Hazards&Robots: A Dataset for Visual Anomaly Detection in Robotics**
+
+Authors:
+
+- Dario Mantegazza
+- Alind Xhyra
+- Luca M. Gambardella
+- Alessandro Giusti
+- Jérôme Guzzi
+
+Original resources:
+
+- Zenodo: https://zenodo.org/records/7859211
+- GitHub: https://github.com/idsia-robotics/hazard-detection
+- IDSIA Robotics Lab: https://idsia-robotics.github.io/
+
+### Preprocessed Dataset
+
+A preprocessed version of the Corridor dataset prepared for one-class anomaly detection experiments is available on Kaggle:
+
+https://www.kaggle.com/datasets/rashidrao/robotics-hazards
+
+The dataset has been reorganized into train, validation, and test splits following the standard anomaly detection protocol:
+
+- Training set: normal samples only
+- Validation set: normal samples for threshold calibration
+- Test set: normal and anomalous samples
+
+All credit for the original data belongs to the Hazards&Robots authors.
+
+## Model Architecture
+
+The checkpoint is based on a VAE-GAN architecture consisting of:
+
+- Encoder network
+- Variational latent representation
+- Decoder / Generator
+- Adversarial discriminator
+
+The model is trained to learn the distribution of normal robotic corridor scenes and identify anomalies through reconstruction discrepancies.
+
+## Source Code
+
+The complete training and evaluation framework used to produce these checkpoints is available at:
+
+https://github.com/rashidrao-pk/AD_MultiPointThreshold
+
+The repository includes:
+
+- VAE-GAN training pipeline
+- Multi-point thresholding framework
+- Threshold calibration utilities
+- Reconstruction-based anomaly scoring
+- Evaluation scripts
+- Visualization tools
+- Support for multiple anomaly detection baselines
+- Reproducible experiments for visual anomaly detection
+
+Users interested in retraining the models, reproducing the experiments, or extending the framework are encouraged to use the official repository.
+
+## Intended Use
+
+This model is intended for:
+
+- Visual anomaly detection
+- One-class classification
+- Explainable anomaly detection research
+- Industrial safety monitoring
+- Robotics safety applications
+- Benchmarking anomaly detection algorithms
+
+## Citation
+
+If you use this model, please cite both the original Hazards&Robots dataset paper and any associated publications describing this VAE-GAN implementation.
+
+### Original Dataset Citation
+
+Mantegazza, D., Xhyra, A., Gambardella, L. M., Giusti, A., & Guzzi, J. (2023).
+
+*Hazards&Robots: A Dataset for Visual Anomaly Detection in Robotics.*
+
+Data in Brief, 49, 109264.
+
+DOI: https://doi.org/10.1016/j.dib.2023.109264
+
+## Acknowledgements
+
+This repository redistributes trained model weights only. The original dataset remains the intellectual property of the Hazards&Robots authors and is distributed under its original license. We gratefully acknowledge their contribution to the anomaly detection and robotics research community.

checkpoints/model_corridor_64.pt

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:efac80da96881aa8ca6b7f95d01e5c4437a554e140249f8700ae16388fd9025a
+size 243879426