Image Captioning on COCO 2017 (test)
1.16CIDErOriginal
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| OriginalBackbone=LLaVA 1.5-13B, Pruning Category=Original, TFLOP (ratio %)=6.281 (100.00), Calibration used=false, Finetuning used=false2026.02 | 1.16 | — | — | |
| OriginalBackbone=LLaVA 1.5-7B, Pruning Category=Original, TFLOP (ratio %)=3.228 (100.00), Calibration used=false, Finetuning used=false2026.02 | 1.1 | — | — | |
| OTPruneBackbone=LLaVA 1.5-13B, Pruning Category=One-shot, TFLOP (ratio %)=1.002 (15.71), Calibration used=false, Finetuning used=false2026.02 | 1.03 | 1.27 | 1.18 | |
| OTPruneBackbone=LLaVA 1.5-7B, Pruning Category=One-shot, TFLOP (ratio %)=0.512 (15.63), Calibration used=false, Finetuning used=false2026.02 | 1 | 2.36 | 1.36 | |
| M3Backbone=LLaVA 1.5-7B, Pruning Category=Finetuning, TFLOP (ratio %)=0.512 (15.63), Calibration used=false, Finetuning used=true2026.02 | 1 | 2.68 | — | |
| DivPruneBackbone=LLaVA 1.5-13B, Pruning Category=One-shot, TFLOP (ratio %)=1.002 (15.71), Calibration used=false, Finetuning used=false2026.02 | 1 | 2.14 | 1.91 | |
| OriginalBackbone=LLaVA 1.6-7B, Pruning Category=Original, TFLOP (ratio %)=11.849 (100.00), Calibration used=false, Finetuning used=false2026.02 | 1 | — | — | |
| DivPruneBackbone=LLaVA 1.5-7B, Pruning Category=One-shot, TFLOP (ratio %)=0.512 (15.63), Calibration used=false, Finetuning used=false2026.02 | 0.96 | 3.23 | 1.64 | |
| PruMerge-LoRABackbone=LLaVA 1.5-7B, Pruning Category=Finetuning, TFLOP (ratio %)=Variable, Calibration used=false, Finetuning used=true2026.02 | 0.96 | 3.68 | — | |
| DivPrune*Backbone=LLaVA 1.5-13B, Pruning Category=Adaptive, TFLOP (ratio %)=Variable, Calibration used=false, Finetuning used=false2026.02 | 0.94 | 3.14 | — | |
| DivPrune*Backbone=LLaVA 1.5-7B, Pruning Category=Adaptive, TFLOP (ratio %)=Variable, Calibration used=false, Finetuning used=false2026.02 | 0.91 | 4.73 | — | |
| FitPruneBackbone=LLaVA 1.5-7B, Pruning Category=Finetuning, TFLOP (ratio %)=0.513 (15.65), Calibration used=true, Finetuning used=false2026.02 | 0.9 | 5.59 | — | |
| DivPruneBackbone=LLaVA 1.6-7B, Pruning Category=One-shot, TFLOP (ratio %)=1.266 (10.79), Calibration used=false, Finetuning used=false2026.02 | 0.89 | 1.77 | 1.77 | |
| OTPruneBackbone=LLaVA 1.6-7B, Pruning Category=One-shot, TFLOP (ratio %)=1.266 (10.79), Calibration used=false, Finetuning used=false2026.02 | 0.89 | 1.23 | 1.23 | |
| PruMergeBackbone=LLaVA 1.5-13B, Pruning Category=Adaptive, TFLOP (ratio %)=Variable, Calibration used=true, Finetuning used=false2026.02 | 0.8 | 3.64 | — | |
| PruMergeBackbone=LLaVA 1.5-7B, Pruning Category=Adaptive, TFLOP (ratio %)=Variable, Calibration used=false, Finetuning used=false2026.02 | 0.77 | 6.09 | — | |
| FastVBackbone=LLaVA 1.5-13B, Pruning Category=One-shot, TFLOP (ratio %)=1.003 (15.73), Calibration used=false, Finetuning used=false2026.02 | 0.38 | 4.82 | 2.91 | |
| VTWBackbone=LLaVA 1.5-13B, Pruning Category=One-shot, TFLOP (ratio %)=1.030 (16.16), Calibration used=false, Finetuning used=false2026.02 | 0.08 | 6 | 4 | |
| FastVBackbone=LLaVA 1.5-7B, Pruning Category=One-shot, TFLOP (ratio %)=0.514 (15.69), Calibration used=false, Finetuning used=false2026.02 | 0.06 | 8.41 | 3.59 | |
| VTWBackbone=LLaVA 1.6-7B, Pruning Category=One-shot, TFLOP (ratio %)=1.318 (11.23), Calibration used=false, Finetuning used=false2026.02 | 0.06 | 3.82 | 3.82 | |
| FastVBackbone=LLaVA 1.6-7B, Pruning Category=One-shot, TFLOP (ratio %)=1.327 (11.30), Calibration used=false, Finetuning used=false2026.02 | 0.06 | 3.18 | 3.18 | |
| VTWBackbone=LLaVA 1.5-7B, Pruning Category=One-shot, TFLOP (ratio %)=0.603 (18.46), Calibration used=false, Finetuning used=false2026.02 | 0.05 | 8.23 | 3.41 |