Generalized Category Discovery on CUB
90.7Accuracy (All)Hyp-SelEx
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| Hyp-SelExBackbone=DINOv22025.04 | 90.7 | 85.3 | 93.4 | |
| Hyp-SelExVenue=CVPR’25, Backbone=DINOv22025.09 | 90.7 | 85.3 | 93.4 | |
| PartCo-SelExVenue=Ours, Backbone=DINOv22025.09 | 90.6 | 84.5 | 93.2 | |
| NCGCD-SelExVenue=NeurIPS’25, Backbone=DINOv22025.09 | 87.9 | 85.3 | 89.2 | |
| SelExBackbone=DINOv22025.04 | 87.4 | 85.1 | 88.5 | |
| SelExVenue=ECCV’24, Backbone=DINOv22025.09 | 87.4 | 85.1 | 88.5 | |
| ConGCD-SelExVenue=ICCV’25, Backbone=DINOv22025.09 | 86.3 | 87.4 | 85.8 | |
| PartCo-SelExVenue=Ours, Backbone=DINOv32025.09 | 86.1 | 84.4 | 87 | |
| PartCo-FlipClassVenue=Ours, Backbone=DINOv22025.09 | 85.2 | 86.3 | 84.7 | |
| SelEx-CLIP + AL-GCDBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 84.1 | 79.7 | 86.3 | |
| SelExVenue=ECCV’24, Backbone=DINOv32025.09 | 83.5 | 78.1 | 86.2 | |
| PartCo-SPTNetVenue=Ours, Backbone=DINOv22025.09 | 82.6 | 82.3 | 81.8 | |
| MOSVenue=CVPR’25, Backbone=DINOv22025.09 | 81.1 | 82.1 | 80.6 | |
| PartCo-SimGCDVenue=Ours, Backbone=DINOv22025.09 | 81.1 | 82.4 | 80.5 | |
| ZS CLIP H/14Textual Source=Class Names2026.02 | 80.6 | 81.7 | 80.1 | |
| Hyp-SelExBackbone=DINO2025.04 | 79.8 | 75.8 | 81.8 | |
| Hyp-SelExBackbone=DINOv12026.02 | 79.8 | 75.8 | 81.8 | |
| FlipClassVenue=NeurIPS’24, Backbone=DINOv22025.09 | 79.3 | 80.7 | 78.5 | |
| SpectralGCDTextual Source=Tags2026.02 | 79.2 | 80.4 | 78.5 | |
| PartCo-SimGCDVenue=Ours, Backbone=DINOv32025.09 | 78.8 | 83.4 | 76.6 | |
| RLCDVenue=ICML’25, Backbone=DINOv22025.09 | 78.7 | 79.5 | 78.3 | |
| AllGCDVenue=ICCV’25, Backbone=DINOv22025.09 | 78.4 | 82.8 | 76.2 | |
| CiPRBackbone=DINOv22025.04 | 78.3 | 73.4 | 80.8 | |
| SSR2-GCDBackbone=CLIP2026.02 | 78.3 | 78.5 | 78.2 | |
| SSR2-GCD2026.02 | 78.3 | 78.5 | 78.2 | |
| CiPRVenue=TMLR, Backbone=DINOv22025.09 | 78.3 | 73.4 | 80.8 | |
| Hyp-SimGCDBackbone=DINOv22025.04 | 77.6 | 77.9 | 77.4 | |
| PartGCDVenue=TMM, Backbone=DINOv22025.09 | 77.6 | 80.6 | 76.1 | |
| Hyp-SimGCDVenue=CVPR’25, Backbone=DINOv22025.09 | 77.6 | 77.9 | 77.4 | |
| DebGCDVenue=ICLR’25, Backbone=DINOv22025.09 | 77.5 | 80.8 | 75.8 | |
| GETTextual Source=InversionNet2026.02 | 77 | 78.1 | 76.4 | |
| GETBackbone=CLIP2026.02 | 77 | 78.1 | 76.4 | |
| GET2026.02 | 77 | 78.1 | 76.4 | |
| GETBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 77 | 78.1 | 76.4 | |
| GETEncoder=Image & Text, Backbone=CLIP2025.09 | 77 | 78.1 | 76.4 | |
| SEALVenue=NeurIPS’25, Backbone=DINOv22025.09 | 76.7 | 78.3 | 75.9 | |
| TextGCDTextual Source=Tags + Attributes2026.02 | 76.6 | 80.6 | 74.7 | |
| TextGCDBackbone=CLIP2026.02 | 76.6 | 80.6 | 74.7 | |
| TextGCD2026.02 | 76.6 | 80.6 | 74.7 | |
| TextGCDEncoder=Image & Text, Backbone=CLIP2025.09 | 76.6 | 80.6 | 74.7 | |
| AFGCDVenue=ICCV’25, Backbone=DINOv22025.09 | 76.5 | 77.3 | 76 | |
| SPTNetBackbone=DINOv22025.04 | 76.3 | 79.5 | 74.6 | |
| SPTNetVenue=ICLR’24, Backbone=DINOv22025.09 | 76.3 | 79.5 | 74.6 | |
| SimGCDVenue=ICCV’23, Backbone=DINOv32025.09 | 75.9 | 83.8 | 72 | |
| ProtoGCDVenue=TPAMI, Backbone=DINOv22025.09 | 75.7 | 81.5 | 72.9 | |
| Hyp-GCDBackbone=DINOv22025.04 | 75.6 | 75.1 | 75.9 | |
| APLVenue=CVPR’25, Backbone=DINOv22025.09 | 75.1 | 79.1 | 73.2 | |
| SimGCD-CLIP + AL-GCDBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 74.7 | 79.6 | 72.3 | |
| SelEx-CLIPBackbone=CLIP, Ground-truth number of categories K=Given, reproduced=true2026.03 | 74.2 | 69.5 | 76.5 | |
| µGCDBackbone=DINOv22025.04 | 74 | 75.9 | 73.1 | |
| µGCDVenue=NeurIPS’23, Backbone=DINOv22025.09 | 74 | 75.9 | 73.1 | |
| CMS-CLIP + AL-GCDBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 73.8 | 79 | 71.2 | |
| CMS-CLIP + AL-GCDBackbone=CLIP, Ground-truth number of categories K=Not Given2026.03 | 73.7 | 79.1 | 71 | |
| SelExBackbone=DINO2025.04 | 73.6 | 75.3 | 72.8 | |
| SelEx2026.02 | 73.6 | 75.3 | 72.8 | |
| SelExBackbone=DINOv12026.02 | 73.6 | 75.3 | 72.8 | |
| SelExBackbone=DINO, Ground-truth number of categories K=Given2026.03 | 73.6 | 75.3 | 72.8 | |
| PartCo-SimGCD (Ours)Encoder=Image, Backbone=CLIP2025.09 | 73.4 | 80.2 | 70 | |
| GCDBackbone=DINOv22025.04 | 71.9 | 71.2 | 72.3 | |
| GCDVenue=CVPR’22, Backbone=DINOv22025.09 | 71.9 | 71.2 | 72.3 | |
| SimGCDEncoder=Image, Backbone=CLIP2025.09 | 71.7 | 76.5 | 69.4 | |
| SimGCDBackbone=DINOv22025.04 | 71.5 | 78.1 | 68.3 | |
| SimGCDVenue=ICCV’23, Backbone=DINOv22025.09 | 71.5 | 78.1 | 68.3 | |
| CPTBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 70.1 | 73.5 | 68.4 | |
| CPTEncoder=Image & Text, Backbone=CLIP2025.09 | 70.1 | 73.5 | 68.4 | |
| RLCDBackbone=DINO, Ground-truth number of categories K=Given2026.03 | 70 | 79.1 | 65.4 | |
| TextGCD*Textual Source=Tags2026.02 | 69.9 | 74.9 | 67.3 | |
| k-meansBackbone=DINOv32025.09 | 69.8 | 64.7 | 72.3 | |
| MOSVenue/Year=our2025.03 | 69.6 | 72.3 | 68.2 | |
| SimGCD-CLIPBackbone=CLIP, Ground-truth number of categories K=Given, reproduced=true2026.03 | 69.6 | 75.8 | 66.5 | |
| InfoSieveVenue/Year=NIPS/20232025.03 | 69.4 | 77.9 | 65.2 | |
| InfoSieveBackbone=DINO2025.04 | 69.4 | 77.9 | 65.2 | |
| InfoSieveBackbone=ViT-B/162025.12 | 69.4 | 77.9 | 65.2 | |
| Sharpness-aware Dynamic Anchor Selection (Ours)Backbone=ViT-B/162025.12 | 69 | 69.1 | 68.9 | |
| OursKnown C=GT (200)2025.12 | 69 | 69.1 | 68.9 | |
| GCAVenue/Year=WACV/20242025.03 | 68.8 | 73.4 | 66.6 | |
| CMSBackbone=DINO2025.04 | 68.2 | 76.5 | 64 | |
| CMSBackbone=ViT-B/162025.12 | 68.2 | 76.5 | 64 | |
| CMSBackbone=DINO, Ground-truth number of categories K=Given2026.03 | 68.2 | 76.5 | 64 | |
| k-meansBackbone=DINOv22025.09 | 67.6 | 60.6 | 71.1 | |
| RPC2026.05 | 67.1 | 73.2 | 64.8 | |
| InfoSculpt2026.01 | 66.8 | 72 | 59.9 | |
| ActiveGCD2026.01 | 66.6 | 66.5 | 66.7 | |
| OursKnown C=Est. (231)2025.12 | 66.6 | 67 | 66.4 | |
| DebGCD2026.02 | 66.3 | 71.8 | 63.5 | |
| DebGCD2026.05 | 66.3 | 71.8 | 63.5 | |
| SPTNetVenue/Year=ICLR/20242025.03 | 65.8 | 68.8 | 65.1 | |
| SPTNetBackbone=DINO2025.04 | 65.8 | 68.8 | 65.1 | |
| SPTNetBackbone=ViT-B/162025.12 | 65.8 | 68.8 | 65.1 | |
| SPTNetTextual Source=✗2026.02 | 65.8 | 68.8 | 65.1 | |
| SPTNetBackbone=DINOv12026.02 | 65.8 | 68.8 | 65.1 | |
| CMS-CLIPBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 65.8 | 75.3 | 61.1 | |
| SPTNet2026.05 | 65.8 | 68.8 | 65.1 | |
| μGCDVenue/Year=NIPS/20232025.03 | 65.7 | 68 | 64.6 | |
| µGCDBackbone=DINO2025.04 | 65.7 | 68 | 64.6 | |
| µGCDBackbone=ViT-B/162025.12 | 65.7 | 68 | 64.6 | |
| µGCDKnown C=GT (200)2025.12 | 65.7 | 68 | 64.6 | |
| CMS-CLIPBackbone=CLIP, Ground-truth number of categories K=Not Given2026.03 | 65.6 | 74 | 61.3 | |
| ProtoGCD2026.01 | 65.2 | 68.5 | 60.5 | |
| RSSBackbone=ViT-B/162025.12 | 65.1 | 66.2 | 64.5 |