Camera Pose Estimation on CO3D v2
97.31AUC@30FP16
Evaluation results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| FP16Bitwidth (W/A)=16/162026.05 | 97.31 | — | 94.62 | 84.62 | 76.3 | — | — | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.1, CDF Thr.=0.97, Sparsity=0.08, Backbone=pi^32026.03 | 97.28 | — | — | — | — | — | 5.4822 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.3, CDF Thr.=0.97, Sparsity=0.11, Backbone=pi^32026.03 | 97.26 | — | — | — | — | — | 5.6254 | — | — | — | |
| SparseVGGTSparse Ratio=0.1, CDF Thr.=0.97, Sparsity=0.08, Backbone=pi^32026.03 | 97.22 | — | — | — | — | — | 5.581 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.4, CDF Thr.=0.95, Sparsity=0.16, Backbone=pi^32026.03 | 97.22 | — | — | — | — | — | 5.7517 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.5, CDF Thr.=0.97, Sparsity=0.13, Backbone=pi^32026.03 | 97.16 | — | — | — | — | — | 5.6833 | — | — | — | |
| SparseVGGTSparse Ratio=0.3, CDF Thr.=0.97, Sparsity=0.11, Backbone=pi^32026.03 | 97.13 | — | — | — | — | — | 5.7365 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.5, CDF Thr.=0.88, Sparsity=0.28, Backbone=pi^32026.03 | 97.12 | — | — | — | — | — | 6.0526 | — | — | — | |
| SparseVGGTSparse Ratio=0.5, CDF Thr.=0.97, Sparsity=0.13, Backbone=pi^32026.03 | 97.09 | — | — | — | — | — | 5.808 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.6, CDF Thr.=0.93, Sparsity=0.22, Backbone=pi^32026.03 | 97.08 | — | — | — | — | — | 5.9403 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.5, CDF Thr.=0.90, Sparsity=0.25, Backbone=pi^32026.03 | 97.08 | — | — | — | — | — | 5.9556 | — | — | — | |
| SparseVGGTSparse Ratio=0.4, CDF Thr.=0.95, Sparsity=0.16, Backbone=pi^32026.03 | 97.04 | — | — | — | — | — | 5.9064 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.5, CDF Thr.=0.75, Sparsity=0.42, Backbone=pi^32026.03 | 97.01 | — | — | — | — | — | 6.4302 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.5, CDF Thr.=0.80, Sparsity=0.37, Backbone=pi^32026.03 | 97 | — | — | — | — | — | 6.2508 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.5, CDF Thr.=0.85, Sparsity=0.32, Backbone=pi^32026.03 | 96.99 | — | — | — | — | — | 6.1476 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.5, CDF Thr.=0.70, Sparsity=0.46, Backbone=pi^32026.03 | 96.9 | — | — | — | — | — | 6.5217 | — | — | — | |
| SparseVGGTSparse Ratio=0.6, CDF Thr.=0.93, Sparsity=0.22, Backbone=pi^32026.03 | 96.88 | — | — | — | — | — | 6.167 | — | — | — | |
| FGQBitwidth (W/A)=4/82026.05 | 96.86 | — | 93.71 | 81.85 | 72.3 | — | — | — | — | — | |
| QuantVGGTBitwidth (W/A)=4/82026.05 | 96.84 | — | 93.63 | 81.27 | 71.17 | — | — | — | — | — | |
| SparseVGGTSparse Ratio=0.5, CDF Thr.=0.90, Sparsity=0.25, Backbone=pi^32026.03 | 96.83 | — | — | — | — | — | 6.2365 | — | — | — | |
| FlatQuantBitwidth (W/A)=4/82026.05 | 96.74 | — | 93.51 | 80.51 | 71.91 | — | — | — | — | — | |
| SparseVGGTSparse Ratio=0.5, CDF Thr.=0.88, Sparsity=0.28, Backbone=pi^32026.03 | 96.73 | — | — | — | — | — | 6.424 | — | — | — | |
| QuaRotBitwidth (W/A)=4/82026.05 | 96.72 | — | 93.43 | 80.98 | 70.32 | — | — | — | — | — | |
| SparseVGGTSparse Ratio=0.5, CDF Thr.=0.85, Sparsity=0.32, Backbone=pi^32026.03 | 96.69 | — | — | — | — | — | 6.5726 | — | — | — | |
| SparseVGGTSparse Ratio=0.5, CDF Thr.=0.80, Sparsity=0.37, Backbone=pi^32026.03 | 96.52 | — | — | — | — | — | 6.856 | — | — | — | |
| SparseVGGTSparse Ratio=0.5, CDF Thr.=0.75, Sparsity=0.42, Backbone=pi^32026.03 | 96.46 | — | — | — | — | — | 7.0518 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.6, CDF Thr.=0.60, Sparsity=0.56, Backbone=pi^32026.03 | 96.35 | — | — | — | — | — | 7.3988 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.7, CDF Thr.=0.70, Sparsity=0.51, Backbone=pi^32026.03 | 96.31 | — | — | — | — | — | 7.4775 | — | — | — | |
| SparseVGGTSparse Ratio=0.5, CDF Thr.=0.70, Sparsity=0.46, Backbone=pi^32026.03 | 96.27 | — | — | — | — | — | 7.3248 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.6, CDF Thr.=0.40, Sparsity=0.60, Backbone=pi^32026.03 | 96.27 | — | — | — | — | — | 7.5568 | — | — | — | |
| RTNBitwidth (W/A)=4/82026.05 | 96.19 | — | 92.38 | 78.22 | 66.91 | — | — | — | — | — | |
| SparseVGGTSparse Ratio=0.7, CDF Thr.=0.70, Sparsity=0.51, Backbone=pi^32026.03 | 96.08 | — | — | — | — | — | 7.6261 | — | — | — | |
| SparseVGGTSparse Ratio=0.6, CDF Thr.=0.60, Sparsity=0.56, Backbone=pi^32026.03 | 95.96 | — | — | — | — | — | 7.9918 | — | — | — | |
| SparseVGGTSparse Ratio=0.6, CDF Thr.=0.40, Sparsity=0.60, Backbone=pi^32026.03 | 95.73 | — | — | — | — | — | 8.3667 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.8, CDF Thr.=0.50, Sparsity=0.69, Backbone=pi^32026.03 | 95.05 | — | — | — | — | — | 9.2602 | — | — | — | |
| SparseVGGTSparse Ratio=0.8, CDF Thr.=0.50, Sparsity=0.69, Backbone=pi^32026.03 | 94.95 | — | — | — | — | — | 9.5964 | — | — | — | |
| SparseVGGTSparse Ratio=0.8, CDF Thr.=0.40, Sparsity=0.75, Backbone=pi^32026.03 | 94.58 | — | — | — | — | — | 10.1408 | — | — | — | |
| HeSS-Guided SparsificationSparse Ratio=0.8, CDF Thr.=0.40, Sparsity=0.75, Backbone=pi^32026.03 | 94.55 | — | — | — | — | — | 9.4527 | — | — | — | |
| FP16Bitwidth (W/A)=16/162026.05 | 94.41 | — | 88.82 | 68.11 | 51.4 | — | — | — | — | — | |
| FGQBitwidth (W/A)=4/42026.05 | 94.25 | — | 88.87 | 70.25 | 55.75 | — | — | — | — | — | |
| FGQBitwidth (W/A)=4/42026.05 | 93.94 | — | 87.95 | 67.11 | 50.32 | — | — | — | — | — | |
| QuantVGGTBitwidth (W/A)=4/42026.05 | 93.87 | — | 88.4 | 69.73 | 54.19 | — | — | — | — | — | |
| FlatQuantBitwidth (W/A)=4/42026.05 | 92.95 | — | 86.48 | 64.05 | 46.91 | — | — | — | — | — | |
| TAPTQBit-Width=W8A82026.02 | 92.52 | — | 86.2 | 69.33 | 57.23 | — | — | — | — | — | |
| GPTQBit-Width=W8A82026.02 | 92.25 | — | 86.08 | 69.07 | 56.86 | — | — | — | — | — | |
| PTQ4ViTBit-Width=W8A82026.02 | 92.2 | — | 85.92 | 68.77 | 56.35 | — | — | — | — | — | |
| VGGTHardware (H100 GPU)=one H100 GPU, Number of input frames (10 frames)=10 random frames, Optimization method (e.g., Bundle Adjustment)=Bundle Adjustment (BA)2025.03 | 91.8 | 1.8 | — | — | — | — | — | — | — | — | |
| FlatQuantBitwidth (W/A)=4/42026.05 | 91.19 | — | 82.4 | 50.22 | 26.86 | — | — | — | — | — | |
| π³Attention=Full2026.07 | 90.91 | — | — | — | 60.83 | — | — | — | — | — | |
| π³ + BABundle Adjustment=true2026.05 | 90.7 | — | — | — | 58.4 | — | — | 80.3 | — | — | |
| DA3Attention=Full2026.07 | 90.58 | — | — | — | 53.28 | — | — | — | — | — | |
| HD-VGGT2026.03 | 90.4 | — | — | — | — | — | — | 86.5 | — | — | |
| GLUEMAP2026.05 | 90.4 | — | — | — | 56.7 | — | — | 79.9 | — | — | |
| VGGTAttention=Full2026.07 | 90.34 | — | — | — | 58.26 | — | — | — | — | — | |
| ERQBit-Width=W8A82026.02 | 90.14 | — | 81.93 | 55.96 | 39.06 | — | — | — | — | — | |
| SAF3RAttention=Sparse, Base Architecture=π³2026.07 | 90.14 | — | — | — | 53.98 | — | — | — | — | — | |
| VGGTInference Complexity=O(N^2)2026.03 | 89.99 | — | 83.95 | 67.84 | — | — | — | — | — | — | |
| RTNBit-Width=W8A82026.02 | 89.98 | — | 81.9 | 59.14 | 43.8 | — | — | — | — | — | |
| SAF3RAttention=Sparse, Base Architecture=DA32026.07 | 89.73 | — | — | — | 50.42 | — | — | — | — | — | |
| VGGTFLOPs (T)=5.842025.09 | 89.5 | — | 83.2 | 66.1 | 54.9 | — | — | — | — | — | |
| π³2026.05 | 89.5 | — | — | — | 47.3 | — | — | 76.6 | — | — | |
| GLUEMAP†Global motion averaging=true2026.05 | 89.5 | — | — | — | 47.4 | — | — | 76.8 | — | — | |
| BaselineW/A=FP162026.05 | 89.5 | 0.38 | — | — | — | — | — | — | — | — | |
| QuantVGGTW/A=W8A82026.05 | 89.4 | — | — | — | — | — | — | — | — | — | |
| QVGGTW/A=W4A162026.05 | 89.4 | 0.23 | — | — | — | — | — | — | — | — | |
| Aπ³Attention=Sparse2026.07 | 89.27 | — | — | — | 50.69 | — | — | — | — | — | |
| QuantVGGTW/A=W4A162026.05 | 89.2 | — | — | — | — | — | — | — | — | — | |
| SAF3RAttention=Sparse, Base Architecture=VGGT2026.07 | 89.19 | — | — | — | 52.68 | — | — | — | — | — | |
| ZipMapInference Complexity=O(N)2026.03 | 88.76 | — | 81.64 | 62.46 | — | — | — | — | — | — | |
| VGGTHardware (H100 GPU)=one H100 GPU, Number of input frames (10 frames)=10 random frames, Runtime mode (e.g., Feed-Forward)=Feed-Forward2025.03 | 88.2 | 0.2 | — | — | — | — | — | — | — | — | |
| VGGT2026.03 | 88.2 | — | — | — | — | — | — | 83.3 | — | — | |
| π^3Inference Complexity=O(N^2)2026.03 | 87.93 | — | 79.86 | 57.12 | — | — | — | — | — | — | |
| VGGTInput Setting=Streaming, Variant=Windowed2026.05 | 87.92 | — | 81.31 | 64.43 | — | — | — | — | — | — | |
| SmoothQuantW/A=W8A82026.05 | 87.9 | 0.62 | — | — | — | — | — | — | — | — | |
| AVGGTAttention=Sparse2026.07 | 87.78 | — | — | — | 47.1 | — | — | — | — | — | |
| Fastπ³Attention=Sparse2026.07 | 87.56 | — | — | — | 40.49 | — | — | — | — | — | |
| FastDA3Attention=Sparse2026.07 | 87.31 | — | — | — | 43.42 | — | — | — | — | — | |
| QuantVGGTFLOPs (T)=1.402025.09 | 86.9 | — | 78.7 | 57.3 | 43.6 | — | — | — | — | — | |
| VGGT2025.12 | 86.3 | — | 76.8 | — | — | 81.4 | — | — | — | — | |
| VGGT2026.04 | 86.23 | — | — | — | — | — | — | — | 1.1432 | 0.0285 | |
| Co-MeAttention=Sparse2026.07 | 86.15 | — | — | — | 40.02 | — | — | — | — | — | |
| Mamba-VGGTInput Setting=Streaming2026.05 | 85.44 | — | 82.02 | 58.87 | — | — | — | — | — | — | |
| REPQBit-Width=W8A82026.02 | 85.24 | — | 73.73 | 43.79 | 28.27 | — | — | — | — | — | |
| FastVGGTAttention=Sparse2026.07 | 84.84 | — | — | — | 34.73 | — | — | — | — | — | |
| ADA3Attention=Sparse2026.07 | 84.44 | — | — | — | 39.72 | — | — | — | — | — | |
| VGGSfM v2Hardware (H100 GPU)=one H100 GPU, Number of input frames (10 frames)=10 random frames, Optimization method=Bundle Adjustment2025.03 | 83.4 | 10 | — | — | — | — | — | — | — | — | |
| FastVGGT2025.12 | 83.4 | — | 71.9 | — | — | 77.6 | — | — | — | — | |
| FLAREHardware (H100 GPU)=one H100 GPU, Number of input frames (10 frames)=10 random frames, Concurrent work=true2025.03 | 83.3 | 0.5 | — | — | — | — | — | — | — | — | |
| LiteVGGT2025.12 | 83.2 | — | 71.9 | — | — | 77.3 | — | — | — | — | |
| ZipMapmode=streaming2026.03 | 83.12 | — | 72.58 | 45.38 | — | — | — | — | — | — | |
| ZipMapInput Setting=Streaming, Variant=Stream2026.05 | 83.12 | — | 72.58 | 45.38 | — | — | — | — | — | — | |
| SparseVGGTAttention=Sparse2026.07 | 82.83 | — | — | — | 18.59 | — | — | — | — | — | |
| CUT3RHardware (H100 GPU)=one H100 GPU, Number of input frames (10 frames)=10 random frames, Concurrent work=true2025.03 | 82.8 | 0.6 | — | — | — | — | — | — | — | — | |
| FastVGGTFLOPs (T)=1.682025.09 | 82.7 | — | 71.3 | 39.2 | 25.1 | — | — | — | — | — | |
| Fast3RHardware (H100 GPU)=one H100 GPU, Number of input frames (10 frames)=10 random frames, Concurrent work=true2025.03 | 82.5 | 0.2 | — | — | — | — | — | — | — | — | |
| CUT3R2025.12 | 82.5 | — | 73.8 | — | — | 77.3 | — | — | — | — | |
| Sparseπ³Attention=Sparse2026.07 | 82.14 | — | — | — | 15.14 | — | — | — | — | — | |
| MASt3RHardware (H100 GPU)=one H100 GPU, Number of input frames (10 frames)=10 random frames2025.03 | 81.8 | 9 | — | — | — | — | — | — | — | — | |
| MASt3R2026.03 | 81.8 | — | — | — | — | — | — | 78.2 | — | — | |
| QuaRotBitwidth (W/A)=4/42026.05 | 81.69 | — | 64.71 | 22.52 | 9.04 | — | — | — | — | — |