heisen111/timebridge-vits8

TimeBridge ViT-S/8

Pretrained ViT-S/8 checkpoint for TimeBridge: Self-Supervised Video Representation Learning via Start-End Joint Embedding and In-Between Frame Prediction.

This repository provides the official TimeBridge pretrained model checkpoint used for self-supervised video representation learning.

Model

• Backbone: ViT-S/8
• Training epochs: 400
• Framework: PyTorch
• Checkpoint file: timebridge_vits8_400ep.pth

Overview

TimeBridge learns video representations by modeling the temporal evolution between a start frame and an end frame, and reconstructing intermediate frames as a self-supervised training signal.

The learned features are designed for dense video understanding tasks such as:

• video object segmentation
• semantic propagation
• part propagation

Usage

Load the checkpoint in PyTorch:

import torch

ckpt = torch.load("timebridge_vits8_400ep.pth", map_location="cpu")
print(ckpt.keys())

Citation

@InProceedings{Wang_2026_CVPR,
    author    = {Wang, Qin and Morrison, Abigail and Scharr, Hanno and Krajsek, Kai},
    title     = {TimeBridge: Self-Supervised Video Representation Learning via Start-End Joint Embedding and In-Between Frame Prediction},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2026},
    pages     = {39647-39658}
}