archive/meta-score-BGE-VL-v1.5-mmeb.json

{
    "Classification": {
        "ImageNet-1K": 0.78,
        "N24News": 0.812,
        "HatefulMemes": 0.616,
        "VOC2007": 0.928,
        "SUN397": 0.836,
        "Place365": 0.468,
        "ImageNet-A": 0.449,
        "ImageNet-R": 0.771,
        "ObjectNet": 0.54,
        "Country211": 0.172,
        "avg": 0.6372
    },
    "VQA": {
        "OK-VQA": 0.749,
        "A-OKVQA": 0.599,
        "DocVQA": 0.854,
        "InfographicsVQA": 0.47,
        "ChartQA": 0.487,
        "Visual7W": 0.613,
        "ScienceQA": 0.679,
        "VizWiz": 0.493,
        "GQA": 0.728,
        "TextVQA": 0.814,
        "avg": 0.6486
    },
    "Retrieval": {
        "VisDial": 0.846,
        "CIRR": 0.635,
        "VisualNews_t2i": 0.787,
        "VisualNews_i2t": 0.816,
        "MSCOCO_t2i": 0.793,
        "MSCOCO_i2t": 0.755,
        "NIGHTS": 0.698,
        "WebQA": 0.904,
        "OVEN": 0.576,
        "FashionIQ": 0.269,
        "EDIS": 0.904,
        "Wiki-SS-NQ": 0.676,
        "avg": 0.7215833333333332
    },
    "Visual Grounding": {
        "MSCOCO": 0.824,
        "Visual7W-Pointing": 0.912,
        "RefCOCO": 0.911,
        "RefCOCO-Matching": 0.815,
        "avg": 0.8655
    },
    "avg": 0.693861111111111
}