List your company

Visual Question Answering on GQA (ECE)

6.09ECEMixUp
5.60848.859212.1115.3608Oct 27, 2025

Evaluation results

MethodLinks
MixUpVision Token Count=Full, Training=ranking loss2025.106.09
+MixUp^rlVision Token Budget=Full Vision Tokens2025.106.09
MergeMixVision Token Count=Full, Training=ranking loss2025.106.5
+MergeMix^rlVision Token Budget=Full Vision Tokens2025.106.5
CutMixVision Token Count=Full, Training=ranking loss2025.106.74
+CutMix^rlVision Token Budget=Full Vision Tokens2025.106.74
SFT VisionVision Token Count=Full2025.108.52
SFT VisionVision Token Budget=Full Vision Tokens2025.108.52
MergeMixVision Token Count=50%, Training=ranking loss2025.1010.34
+MergeMix^rlVision Token Budget=50% Vision Tokens2025.1010.34
CutMixVision Token Count=50%, Training=ranking loss2025.1010.48
+CutMix^rlVision Token Budget=50% Vision Tokens2025.1010.48
+ResizeMix^rlVision Token Budget=25% Vision Tokens2025.1010.66
+CutMix^rlVision Token Budget=25% Vision Tokens2025.1012.23
ResizeMixVision Token Count=Full, Training=ranking loss2025.1012.53
+ResizeMix^rlVision Token Budget=Full Vision Tokens2025.1012.53
+MergeMix^rlVision Token Budget=25% Vision Tokens2025.1012.55
ResizeMixVision Token Count=50%, Training=ranking loss2025.1012.6
+ResizeMix^rlVision Token Budget=50% Vision Tokens2025.1012.6
+MixUp^rlVision Token Budget=25% Vision Tokens2025.1012.97
SFT VisionVision Token Budget=25% Vision Tokens2025.1013.32
MixUpVision Token Count=50%, Training=ranking loss2025.1013.4
+MixUp^rlVision Token Budget=50% Vision Tokens2025.1013.4
Baseline2025.1014.57
BaselineVision Token Budget=Full (Implied)2025.1014.57
SFT VisionVision Token Count=50%2025.1018.13
SFT VisionVision Token Budget=50% Vision Tokens2025.1018.13