README.md

metadata

license: cc-by-nc-sa-4.0
library_name: transformers
pipeline_tag: image-classification

SPTNet: An Efficient Alternative Framework for Generalized Category Discovery with Spatial Prompt Tuning (ICLR 2024)

This repository contains the model described in https://arxiv.org/abs/2403.13684.

Code: https://github.com/Visual-AI/SPTNet

SPTNet: An Efficient Alternative Framework for Generalized Category Discovery with Spatial Prompt Tuning
By Hongjun Wang, Sagar Vaze, and Kai Han.

[05.2024] We update the results of SPTNet with DINOv2 on CUB, please check our latest version in Arxiv

	All	Old	New
CUB (DINO)	65.8	68.8	65.1
CUB (DINOv2)	76.3	79.5	74.6

Results

Generic results:

	All	Old	New
CIFAR-10	97.3	95.0	98.6
CIFAR-100	81.3	84.3	75.6
ImageNet-100	85.4	93.2	81.4

Fine-grained results:

	All	Old	New
CUB	65.8	68.8	65.1
Stanford Cars	59.0	79.2	49.3
FGVC-Aircraft	59.3	61.8	58.1
Herbarium19	43.4	58.7	35.2

Citing this work

If you find this repo useful for your research, please consider citing our paper:

@inproceedings{wang2024sptnet,
    author    = {Wang, Hongjun and Vaze, Sagar and Han, Kai},
    title     = {SPTNet: An Efficient Alternative Framework for Generalized Category Discovery with Spatial Prompt Tuning},
    booktitle = {International Conference on Learning Representations (ICLR)},
    year      = {2024}
}