vico-uoe/oddoneoutad

How vico-uoe/oddoneoutad works

Odd-One-Out is a multi-view 3D scene reconstruction and anomaly-detection system. Its documented workflow loads scene images and camera data, learns a feature volume and object correspondences in Stage 1, then trains or evaluates anomaly classification in Stage 2; demo inference also exposes reconstruction and classification outputs. The model’s encoder, volume decoder, matching transformer and volume renderers form the main internal layer. The README also documents scene visualizations and a browser-based data viewer. Some stage details and evaluation wiring are documented but their source was not sampled, so those links cite the README.

Main entry points: model.py Odd-One-Out model · train.py Training runner · demo.py Inference demo · misc.py Object and box analysis

Components 18

Data and workflows

Scene dataset data/dataset_e2e.py
Model configs configs
Training runner train.py
3D correspondence train.py
Anomaly training train.py
Model evaluation test.py

3D learning network

Odd-One-Out model models/model.py
2D feature encoder models/encoder.py
3D volume decoder models/decoder.py
3D network layers models/layers.py
Feature matching models/xformer.py
Volume render losses models/render.py

Scene analysis

Object and box analysis utils/misc.py
Feature visualization utils/pca.py

Inference and visualization

Inference demo demo.py
Scene data viewer visualize_data.py
Viewer helpers utils/visualize_data.py

Other components

Researcher

Connections 23

Researcher → Training runner: starts training README.md
Researcher → Model evaluation: runs evaluation README.md
Researcher → Inference demo: runs inference README.md
Researcher → Scene data viewer: opens viewer README.md
Training runner → Scene dataset: loads batches train.py
Training runner → Odd-One-Out model: trains train.py
Training runner → Model configs: reads config README.md
Training runner → 3D correspondence: runs stage README.md
Training runner → Anomaly training: runs stage README.md
3D correspondence → Anomaly training: provides checkpoint README.md
Model evaluation → Odd-One-Out model: evaluates README.md
Inference demo → Odd-One-Out model: builds and calls README.md
Odd-One-Out model → 2D feature encoder: uses models/model.py
Odd-One-Out model → 3D volume decoder: uses models/model.py
Odd-One-Out model → Feature matching: uses models/model.py
Odd-One-Out model → Volume render losses: uses models/model.py
3D volume decoder → 3D network layers: uses models/decoder.py
Scene dataset → Object and box analysis: imports helpers data/dataset_e2e.py
Odd-One-Out model → Object and box analysis: uses models/model.py
Odd-One-Out model → Feature visualization: uses models/model.py
Inference demo → Object and box analysis: produces scene analysis README.md
Inference demo → Feature visualization: visualizes features README.md
Scene data viewer → Viewer helpers: uses helpers

More Python diagrams: thehardeep/fenix · tom-the-bomb/discord-games · torch-spyre/torch-spyre · w3f/jamtestvectors · wangzhengpengjay/ai-health-steward

Drawn by GitDiagram on September 30, 2026 from the repository's file tree, README and source files. Read as Markdown ·

How vico-uoe/oddoneoutad works

Main entry points: model.py Odd-One-Out model · train.py Training runner · demo.py Inference demo · misc.py Object and box analysis

Connections 23

Researcher → Training runner: starts training README.md
Researcher → Model evaluation: runs evaluation README.md
Researcher → Inference demo: runs inference README.md
Researcher → Scene data viewer: opens viewer README.md
Training runner → Scene dataset: loads batches train.py
Training runner → Odd-One-Out model: trains train.py
Training runner → Model configs: reads config README.md
Training runner → 3D correspondence: runs stage README.md
Training runner → Anomaly training: runs stage README.md
3D correspondence → Anomaly training: provides checkpoint README.md
Model evaluation → Odd-One-Out model: evaluates README.md
Inference demo → Odd-One-Out model: builds and calls README.md
Odd-One-Out model → 2D feature encoder: uses models/model.py
Odd-One-Out model → 3D volume decoder: uses models/model.py
Odd-One-Out model → Feature matching: uses models/model.py
Odd-One-Out model → Volume render losses: uses models/model.py
3D volume decoder → 3D network layers: uses models/decoder.py
Scene dataset → Object and box analysis: imports helpers data/dataset_e2e.py
Odd-One-Out model → Object and box analysis: uses models/model.py
Odd-One-Out model → Feature visualization: uses models/model.py
Inference demo → Object and box analysis: produces scene analysis README.md
Inference demo → Feature visualization: visualizes features README.md
Scene data viewer → Viewer helpers: uses helpers

Drawn by GitDiagram on September 30, 2026 from the repository's file tree, README and source files. Read as Markdown ·