Dataset Condensation on ImageNet 2015 (train)
3.43FIDGeometry-Aware Dataset Condensation
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Geometry-Aware Dataset CondensationData Budget=0.8% (10K), Eval. Samples=50K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 3.43 | 414.3 | 78 | 28 | |
| CCSData Budget=0.8% (10K), Eval. Samples=50K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 5.45 | 364.9 | 77 | 21 | |
| Geometry-Aware Dataset CondensationData Budget=0.8% (10K), Eval. Samples=10K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 5.76 | 430.2 | 78 | 69 | |
| CCSData Budget=0.8% (10K), Eval. Samples=10K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 7.81 | 373.3 | 77 | 58 | |
| IGDData Budget=0.8% (10K), Eval. Samples=50K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 12.67 | 412.5 | 85 | 10 | |
| IGDData Budget=0.8% (10K), Eval. Samples=10K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 15.92 | 433.9 | 83 | 41 | |
| RDEDData Budget=0.8% (10K), Eval. Samples=50K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 20.9 | 482.7 | 82 | 5 | |
| RDEDData Budget=0.8% (10K), Eval. Samples=10K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 24.1 | 491 | 80 | 25 | |
| EDCData Budget=0.8% (10K), Eval. Samples=50K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 29.4 | 161.7 | 52 | 20 | |
| EDCData Budget=0.8% (10K), Eval. Samples=10K, Backbone=DiT-L/2, Resolution=256x256, Training Iterations=100K, Attachment Phase=D2C2026.06 | 31.7 | 161.5 | 52 | 63 |