Mismatch between result message and result image

I made custom dataset that has 3 classes.

And, I trained hrnet model with pretrained weights.
`python train.py --gpus 0,1,2 --cfg config/ade20k-hrnetv2.yaml`

Then, I validated a one image used training.

Output message is `Mean IoU: 1.0000, Accuracy: 100.00%`

But, when looking result image, it's not.

I printed number of pixcel in intersection and union.
And, number of gray pixcel is 12614.
However, it is must be > 1,000,000, considering result image.

`### intersection: [12614 4295 3970] union [12614 4295 3970]`

I have no idea what is problem.

Reproduce the validation run from train.py using config/ade20k-hrnetv2.yaml and the reported three-class dataset. Compare the inputs used for Mean IoU and accuracy with the result image and the printed intersection/union counts. Done means identifying why the perfect metrics disagree with the visual output and documenting a reproducible correction or confirmed data issue.

Written by the indexing model from the issue text.