about iou

[zhijiejia]

hi feevos, the result in your paper has IOU metric in table. Is the IOU metric for change only ? Or the IOU euqal mIOU for 2 classes

[feevos]

Hi @zhijiejia, It is the IoU metric of the class "change" only, i.e. it does not account for the overlap in the background (no change) class. The code I used for calculating IoU over all test images for LEVIRCD is the following (note that in the inner product, 1 stands for change class, 0 for background, thefore, background does not contribute):

total_intersection = 0.0
total_union = 0.0
for name_pred, name_gt in zip(flnames_inf, flnames_gt):
    print ("Processing file:: {}".format(name_gt))
    with rasterio.open(name_pred,mode='r',driver='GTiff') as src1:
        pred = src1.read((1,2))
    pred = np.argmax(pred,0).astype(np.uint8)
    with rasterio.open(name_gt, driver='png',mode='r') as src:
        mask = src.read(1).astype(np.uint8)
    mask[mask>0]=1

    intersection = (pred*mask).sum()
    union = pred.sum() + mask.sum() - intersection
    
    total_intersection += intersection
    total_union += union

print (total_intersection/total_union)

Regards