Where are testA and testB ?

[nightbreezecool]

Does the test set used in the experiment combine testA and testB? Is the IOU result output by test.py the mean value of testA and testB?

[kkakkkka]

No, it's not. You can make corresponding modifications to the scripts in run_scripts to calculate the IOU of different subsets. Here is a code implementation that may help you better understand the distribution of subsets in different datasets

info = {
    'refcoco': {
        'train': 42404,
        'val': 3811,
        'val-test': 3811,
        'testA': 1975,
        'testB': 1810
    },
    'refcoco+': {
        'train': 42278,
        'val': 3805,
        'val-test': 3805,
        'testA': 1975,
        'testB': 1798
    },
    'refcocog_u': {
        'train': 42226,
        'val': 2573,
        'val-test': 2573,
        'test': 5023
    },
    'refcocog_g': {
        'train': 44822,
        'val': 5000,
        'val-test': 5000
    }
}

[nightbreezecool]

Is the value of MIoU in the article the mean of the IoU of the data sets val, testA and testB?

[nightbreezecool]

Then, is there any part of the project that calculates the loss of the validate set?