复现指数太低

[620ztt]

请问python tools/test.py configs/changer/changer_ex_r18_512x512_40k_levircd.py changer_r18_levir_workdir/latest.pth的latest.pth是指iter_40000.pth
，还是best_mIoU_iter_16000.pth，
使用latest.pth的指标很低，

使用best_mIoU_iter_16000.pth的指标为

这是什么原因

[Camille7911]

Have you solved it? I initially think it’s a label value issue. In the mmseg package, the label mask values are like (0,0,1), but the labels we downloaded are (0,0,255). The author should explain this problem or directly provide properly processed dataset files. Hopefully, the author can pay attention to this issue.

[gl2yx]

Have you solved it? I initially think it’s a label value issue. In the mmseg package, the label mask values are like (0,0,1), but the labels we downloaded are (0,0,255). The author should explain this problem or directly provide properly processed dataset files. Hopefully, the author can pay attention to this issue.

For datasets with labels in the range of [0, 255], it is necessary to add format_seg_map='to_binary' in the dataloader. For example:
train_dataloader = dict(
batch_size=8,
num_workers=4,
persistent_workers=True,
sampler=dict(type='InfiniteSampler', shuffle=True),
dataset=dict(
type='LEVIR_CD_Dataset',
format_seg_map='to_binary', # ←Here
data_root=data_root,
data_prefix=dict(
seg_map_path='train/label',
img_path_from='train/A',
img_path_to='train/B'),
pipeline=train_pipeline))