This repository is the implementation of the paper: ViT2 - Pre-training Vision Transformers for Visual Times Series Forecasting. ViT2 is a framework designed to address generalization & transfer learning limitations of Time-Series-based forecasting models by encoding the time-series to images using GAF and a modified ViT architecture.
python deep-neural-networks computer-vision tensorflow pytorch resnet convolutional-neural-networks darts time-series-forecasting image-encoding vision-transformer nbeats timm temporal-fusion-transformer nhits multi-quantile-regression gramian-angular-fields probalistic-forecasting
- Updated
Apr 8, 2024 - Jupyter Notebook