Generalized Category Discovery on ImageNet 100
93.4All AccuracySpectralGCD
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| SpectralGCDTextual Source=Tags2026.02 | 93.4 | 96.1 | 92 | — | — | — | — | |
| SelEx-CLIP + AL-GCDBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 92.7 | 96.2 | 91.1 | — | — | — | — | |
| SimGCD-CLIP + AL-GCDBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 92.6 | 95.3 | 91.2 | — | — | — | — | |
| Hyp-SelExBackbone=DINOv22025.04 | 92.3 | 96.4 | 90.2 | — | — | — | — | |
| SSR2-GCDBackbone=CLIP2026.02 | 92.1 | 96 | 90.2 | — | — | — | — | |
| SSR2-GCDBackbone=CLIP2026.02 | 92.1 | 96 | 90.2 | — | — | — | — | |
| Hyp-SimGCDBackbone=DINOv22025.04 | 91.9 | 96.2 | 89.8 | — | — | — | — | |
| GETTextual Source=InversionNet2026.02 | 91.7 | 95.7 | 89.7 | — | — | — | — | |
| GETBackbone=CLIP2026.02 | 91.7 | 95.7 | 89.7 | — | — | — | — | |
| GETBackbone=CLIP2026.02 | 91.7 | 95.7 | 89.7 | — | — | — | — | |
| GETBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 91.7 | 95.7 | 89.7 | — | — | — | — | |
| SelEx-CLIPBackbone=CLIP, Ground-truth number of categories K=Given, reproduced=true2026.03 | 91.2 | 93.6 | 90 | — | — | — | — | |
| FlipClassBackbone=DINOv22024.09 | 91 | 96.3 | 88.3 | 93.9 | — | — | — | |
| SelExBackbone=DINOv22025.04 | 90.9 | 96.2 | 88.3 | — | — | — | — | |
| SPTNetBackbone=DINOv22025.04 | 90.1 | 96.1 | 87.1 | — | — | — | — | |
| SimGCDBackbone=DINOv22025.04 | 89.9 | 95.5 | 87.1 | — | — | — | — | |
| SimGCD-CLIPBackbone=CLIP, Ground-truth number of categories K=Given, reproduced=true2026.03 | 89.9 | 94.2 | 87.5 | — | — | — | — | |
| CPTBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 89.2 | 95.2 | 86.1 | — | — | — | — | |
| CiPRBackbone=DINOv22025.04 | 88.2 | 87.6 | 88.5 | — | — | — | — | |
| TextGCDTextual Source=Tags + Attributes2026.02 | 88 | 92.4 | 85.2 | — | — | — | — | |
| TextGCDBackbone=CLIP2026.02 | 88 | 92.4 | 85.2 | — | — | — | — | |
| TextGCDBackbone=CLIP2026.02 | 88 | 92.4 | 85.2 | — | — | — | — | |
| CMS-CLIP + AL-GCDBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 87.6 | 95.9 | 84 | — | — | — | — | |
| AMENDBackbone=DINOv22024.09 | 87.3 | 95.1 | 83.4 | 89.5 | — | — | — | |
| Sharpness-aware Dynamic Anchor Selection (Ours)Backbone=ViT-B/162025.12 | 87.1 | 93.9 | 83.7 | — | — | — | — | |
| RLCDBackbone=DINO, Ground-truth number of categories K=Given2026.03 | 86.9 | 94.2 | 83.2 | — | — | — | — | |
| Hyp-SelExBackbone=DINO2025.04 | 86.8 | 94.6 | 82.8 | — | — | — | — | |
| ZS CLIP H/14Textual Source=Class Names2026.02 | 86.8 | 87.6 | 86.4 | — | — | — | — | |
| Hyp-SelExBackbone=DINOv12026.02 | 86.8 | 94.6 | 82.8 | — | — | — | — | |
| FlipClassBackbone=DINO2024.09 | 86.7 | 94.3 | 82.9 | 90.1 | — | — | — | |
| MTMCVenue=arXiv’252026.01 | 86.7 | 93.1 | 83.6 | — | — | — | — | |
| TextGCD*Textual Source=Tags2026.02 | 86.6 | 92.6 | 83.6 | — | — | — | — | |
| Hyp-SimGCDBackbone=DINO2025.04 | 86.5 | 93.7 | 83 | — | — | — | — | |
| AdaptGCD2026.05 | 86.4 | 94.9 | 82.1 | — | — | — | — | |
| LegoGCDVenue=CVPR’242026.01 | 86.3 | 94.5 | 82.1 | — | — | — | — | |
| LegoGCDBackbone=ViT-B/162025.12 | 86.3 | 94.5 | 82.1 | — | — | — | — | |
| LegoGCD2026.05 | 86.3 | 94.5 | 82.1 | — | — | — | — | |
| LegoGCD2026.07 | 86.3 | 94.5 | 82.1 | — | — | — | — | |
| RPC2026.05 | 86.2 | 93.8 | 82.1 | — | — | — | — | |
| SimGCDBackbone=CLIP2026.02 | 86.1 | 94.5 | 81.9 | — | — | — | — | |
| LegoGCD + CPF-GCDEnhancement=CPF-GCD2026.07 | 86.1 | 93.1 | 82.6 | — | — | — | — | |
| DebGCD2026.02 | 85.9 | 94.3 | 81.6 | — | — | — | — | |
| CMS-CLIPBackbone=CLIP, Ground-truth number of categories K=Given2026.03 | 85.9 | 93.8 | 82 | — | — | — | — | |
| LAGCD2026.05 | 85.9 | 94.9 | 81.4 | — | — | — | — | |
| DebGCD2026.05 | 85.9 | 94.3 | 81.6 | — | — | — | — | |
| ConGCD2026.07 | 85.9 | 93.4 | 82.5 | — | — | — | — | |
| CMS-CLIP + AL-GCDBackbone=CLIP, K (Ground-truth number of categories)=Not given2026.03 | 85.7 | 96.5 | 80.3 | — | — | — | — | |
| CMS-CLIP + AL-GCDBackbone=CLIP, Ground-truth number of categories K=Not Given2026.03 | 85.7 | 96.5 | 80.3 | — | — | — | — | |
| InfoSculpt2026.01 | 85.6 | 93.7 | 77.5 | — | — | — | — | |
| OwMatch+Backbone=ResNet-502024.11 | 85.5 | 91.5 | 79.6 | — | — | — | — | |
| SelEx + CPF-GCDEnhancement=CPF-GCD2026.07 | 85.5 | 94.7 | 80.8 | — | — | — | — | |
| SPTNetBackbone=DINO2024.09 | 85.4 | 93.2 | 81.4 | 88 | — | — | — | |
| SPTNetBackbone=DINO2025.04 | 85.4 | 93.2 | 81.4 | — | — | — | — | |
| RSSBackbone=ViT-B/162025.12 | 85.4 | 92.8 | 81.7 | — | — | — | — | |
| SPTNetBackbone=ViT-B/162025.12 | 85.4 | 93.2 | 81.4 | — | — | — | — | |
| SPTNetTextual Source=✗2026.02 | 85.4 | 93.2 | 81.4 | — | — | — | — | |
| SPTNetBackbone=DINOv12026.02 | 85.4 | 93.2 | 81.4 | — | — | — | — | |
| SPTNet2026.05 | 85.4 | 93.2 | 81.4 | — | — | — | — | |
| SPTNet2026.05 | 85.4 | 93.2 | 81.4 | — | — | — | — | |
| CMS + CPF-GCDEnhancement=CPF-GCD2026.07 | 85.1 | 95.4 | 79.9 | — | — | — | — | |
| CMS-CLIP (CVPR 24)Backbone=CLIP, K (Ground-truth number of categories)=Not given2026.03 | 84.8 | 93.8 | 80.2 | — | — | — | — | |
| CMS-CLIPBackbone=CLIP, Ground-truth number of categories K=Not Given2026.03 | 84.8 | 93.8 | 80.2 | — | — | — | — | |
| CMSBackbone=DINO2024.09 | 84.7 | 95.6 | 79.2 | — | — | — | — | |
| CMSBackbone=DINO2025.04 | 84.7 | 95.6 | 79.2 | — | — | — | — | |
| CMSBackbone=ViT-B/162025.12 | 84.7 | 95.6 | 79.2 | — | — | — | — | |
| CMSBackbone=DINO, Ground-truth number of categories K=Given2026.03 | 84.7 | 95.6 | 79.2 | — | — | — | — | |
| CMS2026.05 | 84.7 | 95.6 | 79.2 | — | — | — | — | |
| SimGCD + CPF-GCDEnhancement=CPF-GCD2026.07 | 84.2 | 92.7 | 80 | — | — | — | — | |
| InfoSieveBackbone=DINO2024.09 | 84.1 | 92.8 | 74.4 | 84.1 | — | — | — | |
| Ours-DINOAssignment Strategy=Greedy-Hungarian, Backbone=DINO2026.03 | 84.1 | 94.3 | 63.4 | — | — | — | — | |
| TALONBackbone=DINO, Matching Strategy=Greedy–Hungarian2026.03 | 84.1 | 94.3 | 63.4 | — | — | — | — | |
| OpenConBackbone=DINO2025.04 | 84 | 93.8 | 81.2 | — | — | — | — | |
| ProtoGCDVenue=TPAMI’252026.01 | 84 | 92.2 | 79.9 | — | — | — | — | |
| ZS CLIP B/16Textual Source=Class Names2026.02 | 84 | 84.9 | 83.5 | — | — | — | — | |
| ClipGCDTextual Source=CC-12M2026.02 | 84 | 95.5 | 78.2 | — | — | — | — | |
| CLIP-GCDBackbone=CLIP2026.02 | 84 | 95.5 | 78.2 | — | — | — | — | |
| ProtoGCD2026.05 | 84 | 92.2 | 79.9 | — | — | — | — | |
| ProtoGCD2026.05 | 84 | 92.2 | 79.9 | — | — | — | — | |
| ProtoGCD2026.07 | 84 | 92.2 | 79.9 | — | — | — | — | |
| SimGCDBackbone=ViT-B/162024.11 | 83.9 | 93.1 | 77.9 | — | — | — | — | |
| Ours-CLIPAssignment Strategy=Greedy-Hungarian, Backbone=CLIP2026.03 | 83.6 | 96.1 | 58.2 | — | — | — | — | |
| TALONBackbone=CLIP, Matching Strategy=Greedy–Hungarian2026.03 | 83.6 | 96.1 | 58.2 | — | — | — | — | |
| PIM + L-RegCategory=InfoMax based2024.10 | 83.4 | 94 | — | — | 78 | — | — | |
| AdaptGCDBackbone=DINO2024.09 | 83.3 | 90.2 | 76.5 | 82.6 | — | — | — | |
| ActiveGCDVenue=CVPR’242026.01 | 83.3 | 90.2 | 76.5 | — | — | — | — | |
| SelEx2026.07 | 83.3 | 94.4 | 77.7 | — | — | — | — | |
| AMENDBackbone=DINO2024.09 | 83.2 | 92.9 | 78.3 | 87 | — | — | — | |
| AMENDBackbone=DINO2025.04 | 83.2 | 92.9 | 78.3 | — | — | — | — | |
| AMEND2026.05 | 83.2 | 92.9 | 78.3 | — | — | — | — | |
| APL2026.07 | 83.2 | 92.6 | 78.5 | — | — | — | — | |
| SimGCD2026.07 | 83.2 | 94 | 77.8 | — | — | — | — | |
| PromptCAL2023.10 | 83.1 | 92.7 | 78.3 | — | — | — | — | |
| PromptCAL-2ndstage=2nd2022.12 | 83.1 | 92.7 | 78.3 | — | — | — | — | |
| PromptCALBackbone=ViT-B/162024.11 | 83.1 | 92.7 | 78.3 | — | — | — | — | |
| PIMCategory=InfoMax based2024.10 | 83.1 | 95.3 | — | — | 77 | — | — | |
| PromptCALBackbone=DINO2025.04 | 83.1 | 92.7 | 78.3 | — | — | — | — | |
| PIMBackbone=DINO2025.04 | 83.1 | 95.3 | 77 | — | — | — | — | |
| SelExBackbone=DINO2025.04 | 83.1 | 93.6 | 77.8 | — | — | — | — | |
| PromptCALBackbone=ViT-B/162025.12 | 83.1 | 92.7 | 78.3 | — | — | — | — | |
| PromptCAL2026.02 | 83.1 | 92.7 | 78.3 | — | — | — | — |