How hadiaeman/anomaly-detection-in-videos-using-spatio-temporal-autoencoders works
This repository presents a notebook-based video anomaly detection project. Its documented workflow takes video sequences, represents them as 10 resized consecutive frames, reconstructs them with a spatio-temporal autoencoder, and uses reconstruction error to identify potential anomalies. The encoder, latent representation, and decoder are shown as distinct model stages. The README describes these relationships, but no notebook source excerpts were available to verify implementation details such as input handling, error thresholds, or result presentation.
Main entry points: Project (1).ipynb Frame Sequences
Components 8
Sequence Preparation
- Frame Sequences
Project (1).ipynb
Autoencoder
- 3D Encoder
Project (1).ipynb - Latent Representation
Project (1).ipynb - 3D Decoder
Project (1).ipynb
Anomaly Detection
- Reconstructed Sequence
Project (1).ipynb - Reconstruction Error
Project (1).ipynb - Potential Anomaly
Project (1).ipynb
Other components
- Video Input
Connections 8
- Video Input → Frame Sequences: forms sequences
- Frame Sequences → 3D Encoder: encodes
- 3D Encoder → Latent Representation: compresses to
- Latent Representation → 3D Decoder: reconstructs from
- 3D Decoder → Reconstructed Sequence: produces
- Frame Sequences → Reconstruction Error: compared with
- Reconstructed Sequence → Reconstruction Error: compared with
- Reconstruction Error → Potential Anomaly: indicates