Shot Transition Detection on Public Data benchmark leaderboard · SOTA2 Research

Shot Transition Detection on Public Data

76.2Segment PrecisionTransVLM

Evaluation Results

Method	Links
TransVLM2026.04		76.2	80.6	78.3	57.4	56.2	56.8	1.58
TransNetV22026.04		72.7	78	75.2	73.1	42.7	52.8	1.87
AutoShot2026.04		70.7	80.4	75.1	70.9	44	53.2	1.78
PySceneDetect (Adaptive)2026.04		65.6	71.6	68.4	64.5	37.2	45.7	1.93
PySceneDetect (Content)2026.04		62.7	75.9	68.6	60.3	38.3	45.3	1.08
PySceneDetect (Hash)2026.04		56.6	77.8	65.4	54.9	41.6	45.7	2.11
Gemini 2.5 Pro2026.04		55.8	52.7	54.2	45.3	36.1	40.1	3.62
Gemini 3 Pro2026.04		52.7	57.3	54.9	46.9	34.3	39.3	2.18
PySceneDetect (Hist)2026.04		45.2	74.1	55.9	41.9	39.8	39.4	1.19
Qwen2-VL Thinking 8B2026.04		45	14.4	21.7	29.7	9.4	14.3	4.25
Qwen2-VL Thinking 4B2026.04		44.9	7.9	13.4	26.5	5.1	8.6	1.5
Qwen2-VL Thinking 32B2026.04		40.3	26	31.5	30.8	16	20.9	1.09
Qwen2-VL Thinking 30B-A3B(MoE)2026.04		37.4	21.7	27.5	29.1	12.7	17.5	0.98
PySceneDetect (Threshold)2026.04		36.4	3.1	5.7	30.6	1.4	2.7	3.08
Qwen2-VL Instruct 32B2026.04		30.9	47.3	37	21.4	27.9	24.1	3.96
Qwen2-VL Instruct 4B2026.04		23.5	8.8	12.4	13.4	17.4	14.8	55
Qwen2-VL Instruct 8B2026.04		22.2	29.7	24.6	13.2	30.7	18.4	34.39
Qwen2-VL Instruct 30B-A3B(MoE)2026.04		21.8	30	24.2	17.1	22.2	19.2	9.61