Add the model card

6d30559 verified about 6 hours ago

4.1 kB

license: other
license_name: vgg-syncnet-research-cc-by
license_link: https://www.robots.ox.ac.uk/~vgg/software/lipsync/
library_name: pytorch
tags:
  - audio-visual
  - lip-sync
  - syncnet

SyncNet `syncnet_v2.model` (unchanged copy)

Status: public copy, uploaded on 2026-11-12 to saracen9/syncnet-v2 by the owner of curu-vision (curu-vision spec 023, research R1.5: the owner's choice of the weights source).

This repository holds an unchanged copy of the file syncnet_v2.model. The copy has the same bytes as the file at the original download URL. curu-vision loads it in the sync port, through its pin syncnet-v2.

The file

Field	Value
File	`syncnet_v2.model`
Size	54,573,114 bytes
SHA-256	`961e8696f888fce4f3f3a6c3d5b3267cf5b343100b238e79b2659bff2c605442`
Downloaded	2026-11-12
Change	None. The file is an unchanged copy of the original file.

The original source

Field	Value
Original download URL	`http://www.robots.ox.ac.uk/~vgg/software/lipsync/data/syncnet_v2.model`, in `download_model.sh` of the code repository. curu-vision pins the same path over HTTPS: `https://www.robots.ox.ac.uk/~vgg/software/lipsync/data/syncnet_v2.model`
Download page	VGG: Out of time: automated lip sync in the wild
Code repository	joonson/syncnet_python
Code revision	`907c0b579c2e2d83f0eae1b2ac9e720cde4e5623` (2026-04-17). The weights file is not in the code repository, so the file itself has no revision. The SHA-256 above identifies it.
Paper	J. S. Chung and A. Zisserman, "Out of time: automated lip sync in the wild", Workshop on Multi-view Lip-reading, ACCV, 2016 (PDF)
Authors	Joon Son Chung and Andrew Zisserman

Licence and terms

Code licence: MIT. LICENSE.md of the code repository starts with "Copyright (c) 2016-present Joon Son Chung." (LICENSE.md).
Terms of the weights: the download page says: "The model can be used for research purposes under Creative Commons Attribution License." It also says: "Please cite [1] if you make use of the software." (download page). The code repository says: "Please cite the paper below if you make use of the software." (README.md).
The owner of curu-vision accepted these terms on 2026-11-12 (curu-vision spec 023, research R1.4: the owner's choice of the weights).

Training data

The released model: not stated. The download page and the code repository do not say which data trained syncnet_v2.model.
The paper: the paper trains SyncNet on BBC television videos. The download page thanks "Rob Cooper and Matt Haynes at BBC Research for help in obtaining the dataset".
The download page also says: "This network can be trained from audio-visual speech datasets, such as VoxCeleb2 or LRS." It does not say that the released model was trained on them.
This copy: it does not change the terms of the original file or of its training data. Check those terms, and the terms of the training datasets, before you use the file.

Use in curu-vision

The sync port of curu-vision (load_sync_detector) loads the file into its own copy of the network of SyncNetModel.py, with the MIT notice.
The integration tier of curu-vision runs the port and SyncNetInstance.py of the code repository on the same crops of the demo clip. Both give the offset 3 and the confidence 9.4839.

Citation

@InProceedings{Chung16a,
  author    = "Chung, J.~S. and Zisserman, A.",
  title     = "Out of time: automated lip sync in the wild",
  booktitle = "Workshop on Multi-view Lip-reading, ACCV",
  year      = "2016",
}

SyncNet syncnet_v2.model (unchanged copy)