Skip to content

Datasets

Anubha Parashar edited this page Aug 10, 2026 · 1 revision

Datasets

IncidentGraph keeps dataset code and experiment logic in the public repository while excluding raw video and other large/restricted data.

MEVA

MEVA is used for the reported real activity-evidence component study. Relevant repository entry points include annotation download, annotated-subset preparation, annotation inventory, manifest building, split repair, training and evaluation scripts.

The reported bounded subset uses 88 downloaded annotated videos and 2,453 activity clips across 28 classes.

MEVID

The codebase contains MEVID download, training and evaluation utilities for re-identification experiments. The real end-to-end IncidentGraph paper claim should not be expanded to MEVID unless the corresponding experiment has actually been run and validated.

MTMMC / WILDTRACK / other adapters

Adapters exist for additional multi-camera sources. Their presence in code indicates integration points, not completed publication experiments.

Public repository policy

Do not commit:

  • raw surveillance videos;
  • private camera feeds;
  • restricted annotations;
  • personal identifiers;
  • large model checkpoints;
  • generated outputs that expose source data;
  • data whose license forbids redistribution.

Dataset-specific licenses and terms remain authoritative even when adapter code is open source.

Clone this wiki locally