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

{
    "Classification": {
        "ImageNet-1K": 0.636,
        "N24News": 0.511,
        "HatefulMemes": 0.51,
        "VOC2007": 0.78,
        "SUN397": 0.722,
        "Place365": 0.431,
        "ImageNet-A": 0.488,
        "ImageNet-R": 0.758,
        "ObjectNet": 0.618,
        "Country211": 0.152,
        "avg": 0.5606000000000001
    },
    "VQA": {
        "OK-VQA": 0.642,
        "A-OKVQA": 0.53,
        "DocVQA": 0.688,
        "InfographicsVQA": 0.319,
        "ChartQA": 0.355,
        "Visual7W": 0.488,
        "ScienceQA": 0.462,
        "VizWiz": 0.515,
        "GQA": 0.755,
        "TextVQA": 0.777,
        "avg": 0.5531
    },
    "Retrieval": {
        "VisDial": 0.701,
        "CIRR": 0.601,
        "VisualNews_t2i": 0.63,
        "VisualNews_i2t": 0.669,
        "MSCOCO_t2i": 0.719,
        "MSCOCO_i2t": 0.63,
        "NIGHTS": 0.598,
        "WebQA": 0.852,
        "OVEN": 0.529,
        "FashionIQ": 0.246,
        "EDIS": 0.807,
        "Wiki-SS-NQ": 0.687,
        "avg": 0.6390833333333334
    },
    "Visual Grounding": {
        "MSCOCO": 0.521,
        "Visual7W-Pointing": 0.789,
        "RefCOCO": 0.74,
        "RefCOCO-Matching": 0.784,
        "avg": 0.7084999999999999
    },
    "avg": 0.6011111111111112
}