Inference Efficiency on MS-COCO
-18.33Sequence Length DeltaGaussian
Evaluation results
| Method | Links | |||
|---|---|---|---|---|
| GaussianModel=GIT2025.06 | -18.33 | -5.21 | -16.68 | |
| NICGSlowdown-BModel=BLIP2025.06 | -9.4 | -9.09 | -3.54 | |
| JPEGModel=Flamingo2025.06 | -7.92 | -0.13 | -5.03 | |
| Verbose-BModel=BLIP2025.06 | -6.98 | 4.24 | -0.84 | |
| GaussianModel=Flamingo2025.06 | -4.15 | -0.16 | -6.92 | |
| NICGSlowdown-BModel=Flamingo2025.06 | -3.54 | 0.19 | -0.22 | |
| Verbose-BModel=Florence2025.06 | -3.01 | -6.06 | -7.88 | |
| Verbose-BModel=Flamingo2025.06 | -2.93 | 5.56 | -0.13 | |
| NICGSlowdown-BModel=Florence2025.06 | -2.56 | -6.59 | -4.54 | |
| JPEGModel=Florence2025.06 | 0.02 | 3.03 | -1.27 | |
| GaussianModel=Florence2025.06 | 1.75 | 3.03 | 0.61 | |
| NICGSlowdown-BModel=GIT2025.06 | 4.27 | 6.67 | 5.89 | |
| Verbose-BModel=GIT2025.06 | 8.23 | 9.31 | 11.22 | |
| GaussianModel=BLIP2025.06 | 18.92 | 18.19 | 26.43 | |
| JPEGModel=GIT2025.06 | 21.47 | 30.09 | 17.25 | |
| JPEGModel=BLIP2025.06 | 28.86 | 27.27 | 39.34 | |
| VLMInferSlowModel=Florence2025.06 | 51.96 | 42.42 | 51.5 | |
| VLMInferSlowModel=BLIP2025.06 | 71.95 | 54.98 | 65.41 | |
| VLMInferSlowModel=GIT2025.06 | 75.93 | 66.67 | 78.59 | |
| VLMInferSlowModel=Flamingo2025.06 | 128.47 | 105.56 | 115.19 |