Visio-Linguistic Compositional Perception on LongCapVLCP 1.0 (test)
93.4Micro Avg ACCEXPERT
Evaluation results
| Method | Links | ||||||
|---|---|---|---|---|---|---|---|
| EXPERTMethod Architecture Category=LLM-based2026.05 | 93.4 | 70.8 | 94.3 | 70.3 | 89.8 | 82.3 | |
| BEiTScoreMethod Architecture Category=Encoder-based, Backbone=L/142026.05 | 90.5 | 68 | 90.7 | 68.5 | 89.8 | 82.3 | |
| VQAScoreMethod Architecture Category=LLM-based, Backbone=CLIP-FlanT5-XXL2026.05 | 87.5 | 77.2 | 86.1 | 72.1 | 93.3 | 91.7 | |
| BEiTScoreMethod Architecture Category=Encoder-based, Backbone=B/322026.05 | 85.4 | 67.3 | 85.5 | 66.6 | 84.9 | 73.7 | |
| FLEURMethod Architecture Category=LLM-based2026.05 | 83.5 | 68.1 | 82.4 | 66.5 | 87.9 | 79.1 | |
| GPT4o (API)Method Architecture Category=LLM-based2026.05 | 77.1 | — | 80.8 | — | 62.3 | — | |
| BEiTScoreMethod Architecture Category=Encoder-based, Backbone=L/14, Training Stage=without Stage 12026.05 | 76.1 | 60.5 | 73.7 | 59.3 | 85.3 | 70.1 | |
| LongCLIP-SMethod Architecture Category=Encoder-based, Backbone=ViT-L2026.05 | 72.6 | 57.6 | 69.2 | 54 | 85.9 | 72.3 | |
| LongCLIP-SMethod Architecture Category=Encoder-based, Backbone=ViT-B2026.05 | 70.9 | 59.7 | 66.7 | 54.5 | 87.4 | 75 | |
| SPECSMethod Architecture Category=Encoder-based2026.05 | 70.5 | 54.6 | 71.8 | 56.8 | 65.7 | 55.8 | |
| BEiTScoreMethod Architecture Category=Encoder-based, Backbone=B/32, Training Stage=without Stage 12026.05 | 66.3 | 55.1 | 64.6 | 53.8 | 72.7 | 60.5 | |
| PAC-SMethod Architecture Category=Encoder-based, Backbone=ViT-L2026.05 | 20.3 | 55.2 | 0 | 50 | 85.8 | 76 | |
| PAC-S++Method Architecture Category=Encoder-based, Backbone=ViT-L2026.05 | 19.9 | 54.8 | 0 | 50 | 86.8 | 74.1 | |
| PAC-SMethod Architecture Category=Encoder-based, Backbone=ViT-B2026.05 | 19.1 | 53.8 | 0 | 50 | 79.2 | 68.7 | |
| PAC-S++Method Architecture Category=Encoder-based, Backbone=ViT-B2026.05 | 19 | 53.5 | 0 | 50 | 79.2 | 67.6 | |
| CLIP-SMethod Architecture Category=Encoder-based, Backbone=ViT-L2026.05 | 17.9 | 55.7 | 0 | 50 | 87.8 | 78.2 | |
| CLIP-SMethod Architecture Category=Encoder-based, Backbone=ViT-B2026.05 | 16.7 | 54.9 | 0 | 50 | 81.7 | 74.7 |