Fine-Grained Sketch-Based Image Retrieval on Chair V2 (test)
89.69Top-1 AccuracyBDCN
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| BDCNTraining Data=BIPED, Parameters=16.3M2023.08 | 89.69 | 98.96 | — | |
| TEEDTraining Data=BIPED, Parameters=58K2023.08 | 88.65 | 100 | — | |
| PiDiNetTraining Data=BIPED, Parameters=710K2023.08 | 87.62 | 100 | — | |
| HOLEF2023.08 | 81.44 | 95.88 | — | |
| Upper-Limit2022.03 | 78.6 | — | 90.3 | |
| Ours-FullTraining Mode=Unlabelled, Backbone=PVT-Large2023.03 | 74.68 | 92.79 | — | |
| B-PKTTraining Mode=Unlabelled, Technique=PKT2023.03 | 73.45 | 91.89 | — | |
| B-RKDTraining Mode=Unlabelled, Technique=RKD2023.03 | 73.02 | 91.78 | — | |
| B-RegressTraining Mode=Unlabelled, Technique=Regression2023.03 | 72.65 | 91.32 | — | |
| B-Edge2SketchTraining Mode=Unlabelled, Technique=Edge2Sketch2023.03 | 72.16 | 91.01 | — | |
| B-Edge-PretrainTraining Mode=Unlabelled, Technique=Edge Pre-training2023.03 | 71.58 | 90.78 | — | |
| Ours-StrongBackbone=PVT-Large, Category=Stronger Baseline2023.03 | 71.22 | 92.18 | — | |
| B-CoAtNetBackbone=CoAtNet, Category=Stronger Baseline2023.03 | 69.68 | 91.78 | — | |
| B-CVTBackbone=CVT, Category=Stronger Baseline2023.03 | 68.42 | 91.21 | — | |
| BDCNTraining Data=BSDS, Parameters=16.3M2023.08 | 68.04 | 98.96 | — | |
| Semi-sup-SN-oursTraining Paradigm=SOTA++, Augmentation=Intra-modal triplet objective2023.03 | 66.86 | 91.12 | — | |
| B-SWINBackbone=SWIN, Category=Stronger Baseline2023.03 | 66.34 | 91.03 | — | |
| StyleMeUp-oursTraining Paradigm=SOTA++, Augmentation=Intra-modal triplet objective2023.03 | 65.85 | 90.84 | — | |
| Proposed (autoregressive sketch mapper)feature extractor=VGG-162023.03 | 65.1 | — | 79.2 | |
| Proposedcomponent=stroke-subset selector2022.03 | 64.8 | — | 79.1 | |
| Ours-RetEvaluation protocol=Zero-shot, Backbone/Variant=Retrieval-based variant2023.03 | 64.31 | 92.6 | — | |
| Ours-RNEvaluation protocol=Zero-shot, Backbone/Variant=ResNet2023.03 | 63.34 | 94.53 | — | |
| StyleMeUp2021.03 | 62.86 | 91.14 | — | |
| StyleMeUpEvaluation protocol=Supervised (trained on target category)2023.03 | 62.86 | 91.14 | — | |
| StyleMeUp2022.03 | 62.8 | — | 79.6 | |
| StyleMeUp2023.03 | 62.8 | — | 79.6 | |
| CrossHier2023.03 | 62.8 | — | 79.1 | |
| Cross-Hier2022.03 | 62.4 | — | 79.1 | |
| B-VGG-19Backbone=VGG-19, Category=Stronger Baseline2023.03 | 61.46 | 89.16 | — | |
| Semi-Sup2022.03 | 60.2 | — | 78.1 | |
| Semi-Sup2023.03 | 60.2 | — | 78.1 | |
| Semi-sup-SNTraining Paradigm=SOTA2023.03 | 60.2 | 90.81 | — | |
| StyleMeUpTraining Paradigm=SOTA2023.03 | 59.86 | 89.64 | — | |
| Linear-Limit2022.03 | 59.4 | — | 77.3 | |
| OnTheFly-oursTraining Paradigm=SOTA++, Augmentation=Intra-modal triplet objective2023.03 | 59.18 | 89.35 | — | |
| Contrastive+Augment2022.03 | 58.8 | — | 77.1 | |
| Jigsaw-SN-oursTraining Paradigm=SOTA++, Augmentation=Intra-modal triplet objective2023.03 | 58.51 | 88.78 | — | |
| B-VGG-16Backbone=VGG-16, Category=Stronger Baseline2023.03 | 58.23 | 88.78 | — | |
| Triplet-RLBackbone=Sketch-A-Net2021.03 | 56.54 | 89.61 | — | |
| TripLet-RLEvaluation protocol=Supervised (trained on target category)2023.03 | 56.54 | 89.61 | — | |
| B-DeITBackbone=DeIT, Category=Stronger Baseline2023.03 | 56.25 | 87.72 | — | |
| Mixed-Jigsaw2022.03 | 56.1 | — | 75.3 | |
| StyleMeUp+Augment2022.03 | 56.1 | — | 76.9 | |
| B-InceptionV3Backbone=InceptionV3, Category=Stronger Baseline2023.03 | 55.41 | 88.21 | — | |
| HOLEF-SN-oursTraining Paradigm=SOTA++, Augmentation=Intra-modal triplet objective2023.03 | 55.23 | 88.61 | — | |
| OnTheFlyTraining Paradigm=SOTA2023.03 | 54.54 | 88.61 | — | |
| CC-Gen2021.03 | 54.21 | 88.23 | — | |
| CC-DGEvaluation protocol=Supervised (trained on target category)2023.03 | 54.21 | 88.23 | — | |
| Augmnt2022.03 | 54.1 | — | 74.6 | |
| B-Meta-SNBackbone=Inception-V3, Optimization=MAML2021.03 | 53.57 | 87.69 | — | |
| Triplet-SN-oursTraining Paradigm=SOTA++, Augmentation=Intra-modal triplet objective2023.03 | 53.48 | 87.91 | — | |
| Triplet-AttnBackbone=Sketch-A-Net2021.03 | 53.41 | 87.56 | — | |
| Jigsaw-SNTraining Paradigm=SOTA2023.03 | 53.41 | 87.56 | — | |
| DSAEvaluation protocol=Supervised (trained on target category)2023.03 | 53.41 | 87.56 | — | |
| Baseline-Siamesestatus=Pre-trained baseline2022.03 | 53.3 | — | 74.3 | |
| D-DVMLDisentanglement=true2021.03 | 52.78 | 85.24 | — | |
| B-Cross-ModalArchitecture=VAE2021.03 | 52.24 | 86.58 | — | |
| Triplet-RL2022.03 | 51.2 | — | 73.8 | |
| Triplet-Attn-HOLEF2022.03 | 50.7 | — | 73.6 | |
| HOLEF-SN2023.03 | 50.7 | — | 73.6 | |
| HOLEF-SNTraining Paradigm=SOTA2023.03 | 50.41 | 86.31 | — | |
| B-SN-GroupBackbone=Inception-V32021.03 | 50.35 | 88.28 | — | |
| B-Basic-SNBackbone=Inception-V32021.03 | 49.58 | 85.41 | — | |
| D-TVAEDisentanglement=true2021.03 | 49.37 | 81.63 | — | |
| B-ResNet-18Backbone=ResNet-18, Category=Stronger Baseline2023.03 | 48.42 | 85.62 | — | |
| B-ResNet-50Backbone=ResNet-50, Category=Stronger Baseline2023.03 | 47.78 | 82.34 | — | |
| Triplet-SNBackbone=Sketch-A-Net2021.03 | 47.65 | 84.24 | — | |
| TripLet-SANEvaluation protocol=Supervised (trained on target category)2023.03 | 47.65 | 84.24 | — | |
| Triplet-SNTraining Paradigm=SOTA2023.03 | 47.45 | 84.32 | — | |
| Triplet-SN2022.03 | 47.4 | — | 71.4 | |
| Triplet-SN2023.03 | 47.4 | — | 71.4 | |
| B-ViTBackbone=ViT, Category=Stronger Baseline2023.03 | 38.71 | 72.65 | — |