Visual Question Answering on CLEVR (val)
100Overall AccuracyXNM-GT
Evaluation Results
| Method | Links | ||||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|
| XNM-GTProgram=GT, #Modules=12, #Param.=0.22M2018.12 | 100 | 100 | 100 | 100 | 100 | 100 | — | — | — | — | |
| XNM-GTProgram=supervised, #Modules=12, #Param.=0.22M2018.12 | 99.9 | 99.9 | 99.8 | 99.9 | 99.8 | 99.9 | — | — | — | — | |
| NS-VQAProgram=supervised, #Modules=122018.12 | 99.8 | 99.7 | 99.9 | 99.9 | 99.8 | 99.8 | — | — | — | — | |
| NSVQAPre-training=Visual Pre-training2022.01 | 99.7 | — | — | — | — | — | — | — | — | — | |
| S-MACPre-training=Visual Pre-training2022.01 | 98.9 | — | — | — | — | — | — | — | — | — | |
| XNM-DetProgram=GT, #Modules=12, #Param.=0.55M2018.12 | 97.9 | 96.2 | 98.8 | 98.1 | 98.7 | 97.8 | — | — | — | — | |
| XNM-DetProgram=supervised, #Modules=12, #Param.=0.55M2018.12 | 97.8 | 96 | 98.6 | 98.1 | 98.7 | 97.8 | — | — | — | — | |
| FiLMPre-training=Visual Pre-training2022.01 | 97.7 | — | — | — | — | — | — | — | — | — | |
| Stack-NMNs**Expert layout=true2018.08 | 96.6 | — | — | — | — | — | — | — | — | — | |
| Object RN**2018.08 | 94.5 | 93.6 | 94.7 | 93.3 | 95.2 | 94.4 | — | — | — | — | |
| VQATransformerPre-training=None2022.01 | 93.6 | — | — | — | — | — | — | — | — | — | |
| Stack-NMNs**Expert layout=false2018.08 | 93 | — | — | — | — | — | — | — | — | — | |
| HumanPre-training=Visual Pre-training2022.01 | 92.6 | — | — | — | — | — | — | — | — | — | |
| VQATransformer + LexSymPre-training=None2022.01 | 92 | — | — | — | — | — | — | — | — | — | |
| VQATransformer + SubstitutePre-training=None2022.01 | 90.8 | — | — | — | — | — | — | — | — | — | |
| SNs (mean)Backbone=Qwen3-VL-4b-Instruct2026.03 | 88.3 | — | — | — | — | — | 88.3 | 87.9 | 90 | 89 | |
| SNs (maj. voting)Backbone=Qwen3-VL-4b-Instruct2026.03 | 88.3 | — | — | — | — | — | 88.3 | 85.3 | 92.5 | 88.7 | |
| Qwen3-VL-4b-Instruct2026.03 | 83.7 | — | — | — | — | — | — | 76.7 | 96.8 | 85.6 | |
| SNs (mean)Backbone=LLaVA-v1.5-7b2026.03 | 51.5 | — | — | — | — | — | 53 | 51 | 75.8 | 61 | |
| SNs (maj. voting)Backbone=LLaVA-v1.5-7b2026.03 | 51.4 | — | — | — | — | — | 53 | 50.9 | 77.5 | 61.4 | |
| LLaVA-v1.5-7b2026.03 | 51.3 | — | — | — | — | — | — | 50.9 | 76.6 | 61.1 |