---
license: mit
library_name: pytorch
tags:
- video-anomaly-detection
- weakly-supervised-learning
- checkpoints
---

# PACT-VAD checkpoints

This repository distributes the validation-selected PACT-VAD model weights used
in the accompanying controlled transfer study.

- Target checkpoints: 3 datasets × 6 conditions × 10 seeds = 180 files
- Source checkpoint: the fixed PreVAD temporal backbone used by the transfer runs
- Selection rule: earliest epoch attaining maximum validation video AUC

Source code, configurations, the canonical metric ledger, and manuscript sources:

https://github.com/tastangh/PACT-VAD

The checkpoint manifest records the dataset, condition, seed, selected epoch,
file size, and SHA-256 digest for every target checkpoint. Dataset videos and
historical feature arrays are not distributed here.