Scene Captioning on COCO RGBP seen scenes (test) benchmark leaderboard · SOTA2 Research

Scene Captioning on COCO RGBP seen scenes (test)

127.58CIDErQwen3-VL-8B-Instruct RGBP

Evaluation Results

Method	Links
Qwen3-VL-8B-Instruct RGBP2026.04		127.58	21.14
ProCap2026.04		78.99	13.83
ProCap2026.04		76.98	13.58
ProCap2026.04		70.27	21.92
ProCap2026.04		69.43	22.01
ProCap2026.04		57.72	10.21
ProCap2026.04		54.37	9.75
Qwen3-VL-8B-Instruct RGBP2026.04		37.81	23.22
ProCap2026.04		36.17	22.75
ProCap2026.04		28.01	24.34
Qwen3-VL-8B-Instruct2026.04		11.56	11.63
Qwen3-VL-2B-Instruct2026.04		10.73	12.47
Qwen3-VL-4B-Instruct2026.04		10.59	11.72
FastVLM-7B2026.04		7.65	7.33
FastVLM-1.5B2026.04		7.21	7.31
FastVLM-0.5B2026.04		7.17	7.43
Qwen3-VL-4B-Instruct2026.04		2.38	11.68
Qwen3-VL-8B-Instruct2026.04		2.38	13.03
Qwen3-VL-2B-Instruct2026.04		2.34	12.17
FastVLM-7B2026.04		2.31	10.31
FastVLM-1.5B2026.04		1.92	8.77
FastVLM-0.5B2026.04		1.72	9.2