Question Answering on Thermo-VL-Bench
89.97Accuracy (RGB)LLaVA-Next-Llama3-8B
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| LLaVA-Next-Llama3-8BLLM backbone=Llama-3-8B, Vision backbone=CLIP ViT-L/142026.05 | 89.97 | 78.67 | 85.89 | 84.73 | |
| Molmo-7B-D-0924LLM backbone=Qwen2-7B, Vision backbone=CLIP ViT-L/142026.05 | 88.67 | 81.2 | 75.79 | 82.51 | |
| Thermo-VLLLM backbone=Qwen2-7B, Vision backbone=CLIP-ViT-L/142026.05 | 88.67 | 82.23 | 88.96 | 86.37 | |
| Qwen2.5-VL-7B-InstructLLM backbone=Qwen2-7B, Vision backbone=QwenViT2026.05 | 83.02 | 86.31 | 76.25 | 82.45 | |
| InternVL 2.5-8BLLM backbone=InternLM2.5-7B, Vision backbone=InternViT-300M-v2.52026.05 | 76.15 | 77.61 | 77.18 | 76.96 |