SPANetV2 Official

August 10, 2026 ยท View on GitHub

๐Ÿ“œ History:

๐Ÿค– It currently includes code and models for the following tasks:

Main results on ImageNet-1K

Please see image_classification for more details.

ModelPretrainResolutionTop-1#Params (M)MACs (G)
SPANeV2-S18-pureImageNet-1K224x22483.4294.2
SPANeV2-S18-hybridImageNet-1K224x22483.9274.2
SPANeV2-S36-pureImageNet-1K224x22484.4448.1
SPANeV2-S36-hybridImageNet-1K224x22484.7418.1
SPANeV2-M36-pureImageNet-1K224x22484.96113.7
SPANeV2-M36-hybridImageNet-1K224x22485.35813.6
SPANeV2-B36-pureImageNet-1K224x22485.010024.3
SPANeV2-B36-hybridImageNet-1K224x22485.610023.9

With the scaled sigmoid for SRF

ModelPretrainResolutionTop-1#Params (M)MACs (G)
SPANeV2-S18-pureImageNet-1K224x22483.5294.2
SPANeV2-S18-hybridImageNet-1K224x224-.-274.2
SPANeV2-S36-pureImageNet-1K224x224-.-448.1
SPANeV2-S36-hybridImageNet-1K224x22484.9418.1

Main results on COCO object detection and instance segmentation

Please see object_detection for more details.

Mask R-CNN 3x

Backbonebox mAPmask mAPMACs (G)
SPANet-S18-pure48.042.9255
SPANet-S18-hybrid49.644.3251

Cascade Mask R-CNN 3x

Backbonebox mAPmask mAPMACs (G)
SPANet-S18-pure51.644.7734
SPANet-S18-hybrid52.845.7729
SPANet-S36-pure53.145.9814
SPANet-S36-hybrid53.846.4806

Main results on ADE20K semantice segmentation

Please see semantic_segmentation for more details.

Semantic FPN

BackbonemIoU#params (M)MACs (G)
SPANet-S18-pure46.73343
SPANet-S18-hybrid47.83148
SPANet-S36-pure47.94864
SPANet-S36-hybrid48.64574

UperNet

BackbonemIoU#params (M)MACs (G)
SPANet-S18-pure48.760930
SPANet-S18-hybrid49.158925
SPANet-S36-pure49.8751012
SPANet-S36-hybrid51.6711004

โญ Cite SPANetV2

If you find this repository useful, please give us stars and use the following BibTeX entry for citation.

@article{yun2026spectral,
  title={Spectral-Adaptive Modulation Networks for Visual Perception},
  author={Yun, Guhnoo and Yoo, Juhan and Kim, Kijung and Lee, Jeongho and Seo, Paul Hongsuck and Kim, Dong Hwan},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2026},
  publisher={IEEE}
}