Novel Object Captioning on NoCaps (val)
85.4CIDEr (In-Domain)Oscar_L
Evaluation Results
| Method | Links | ||||||||
|---|---|---|---|---|---|---|---|---|---|
| Oscar_LTraining Supervision=COCO only without pre-training, Backbone=Large, Inference Constraints=CBS, Optimization Strategy=SCST2020.04 | 85.4 | 11.9 | 84 | 11.7 | 80.3 | 10 | 83.4 | 11.4 | |
| Oscar_BTraining Supervision=COCO only without pre-training, Backbone=Base, Inference Constraints=CBS, Optimization Strategy=SCST2020.04 | 83.4 | 12 | 81.6 | 12 | 77.6 | 10.6 | 81.1 | 11.7 | |
| UpDownTraining Supervision=COCO only without pre-training, Inference Constraints=CBS2020.04 | 80 | 12 | 73.6 | 11.3 | 66.4 | 9.7 | 73.1 | 11.1 | |
| Oscar_BTraining Supervision=COCO only without pre-training, Backbone=Base, Inference Constraints=CBS2020.04 | 80 | 12.1 | 80.4 | 12.2 | 75.3 | 10.6 | 79.3 | 11.9 | |
| Oscar_LTraining Supervision=COCO only without pre-training, Backbone=Large2020.04 | 79.9 | 12.4 | 68.2 | 11.8 | 45.1 | 9.4 | 65.2 | 11.4 | |
| Oscar_BTraining Supervision=COCO only without pre-training, Backbone=Base2020.04 | 79.6 | 12.3 | 66.1 | 11.5 | 45.3 | 9.7 | 63.8 | 11.2 | |
| UpDownTraining Supervision=COCO only without pre-training, Inference Constraints=CBS, Backbone Enhancement=ELMo2020.04 | 79.3 | 12.4 | 73.8 | 11.4 | 71.7 | 9.9 | 74.3 | 11.2 | |
| Oscar_LTraining Supervision=COCO only without pre-training, Backbone=Large, Inference Constraints=CBS2020.04 | 78.8 | 12.2 | 78.9 | 12.1 | 77.4 | 10.5 | 78.6 | 11.8 | |
| UpDownTraining Supervision=COCO only without pre-training2020.04 | 78.1 | 11.6 | 57.7 | 10.3 | 31.3 | 8.3 | 55.3 | 10.1 | |
| ViECap + SeTaPruned %=41.8%, Zero-shot=true2025.03 | 63.9 | — | 67.2 | — | 69.5 | — | 69.3 | — | |
| ViECap + InfoBatchPruned %=23.6%, Zero-shot=true2025.03 | 63.4 | — | 68.2 | — | 70.4 | — | 70.2 | — | |
| ViECap + SeTaPruned %=31.7%, Zero-shot=true2025.03 | 63.4 | — | 69.7 | — | 71.4 | — | 71.5 | — | |
| ViECapPruned %=0%, Zero-shot=true2025.03 | 63.2 | — | 68.8 | — | 70.2 | — | 70.5 | — | |
| ViECap + InfoBatchPruned %=34.1%, Zero-shot=true2025.03 | 61.9 | — | 67.1 | — | 69.6 | — | 69.2 | — | |
| ViECap + SeTaPruned %=50.1%, Zero-shot=true2025.03 | 61.5 | — | 67.1 | — | 69.9 | — | 69.4 | — | |
| ViECap + InfoBatchPruned %=45.9%, Zero-shot=true2025.03 | 60.8 | — | 65.7 | — | 67.5 | — | 67.7 | — |