aimagelab/TransFusion
Official codebase of "Update Your Transformer to the Latest Release: Re-Basin of Task Vectors" - ICML 2025
This tool helps machine learning researchers and practitioners effectively combine or transfer knowledge between different versions of vision models, specifically CLIP and Vision Transformer (ViT) architectures. It takes pre-trained and fine-tuned models as input and outputs a merged model or transferred task vectors, enabling robust evaluation across various image datasets. This is designed for those working with advanced computer vision models.
No commits in the last 6 months.
Use this if you need to integrate knowledge from different pre-trained vision models or transfer specific learned tasks between them, even if their underlying structures aren't perfectly aligned.
Not ideal if you are looking for a simple, off-the-shelf solution for general image classification without needing to merge or transfer knowledge between complex transformer models.
Stars
23
Forks
—
Language
Python
License
—
Category
Last pushed
Jul 30, 2025
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/ml-frameworks/aimagelab/TransFusion"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
Jittor/jittor
Jittor is a high-performance deep learning framework based on JIT compiling and meta-operators.
zhanghang1989/ResNeSt
ResNeSt: Split-Attention Networks
berniwal/swin-transformer-pytorch
Implementation of the Swin Transformer in PyTorch.
NVlabs/FasterViT
[ICLR 2024] Official PyTorch implementation of FasterViT: Fast Vision Transformers with...
ViTAE-Transformer/ViTPose
The official repo for [NeurIPS'22] "ViTPose: Simple Vision Transformer Baselines for Human Pose...