Test problem with DEEPLABV2-RESNET101

[liruihao]

Dear Ali,

I tried to evaluate the segmentation results with DeepLabV2-RESNET101. However, when I used "MatWrite" layer to save the segmentation results to mat files. I found that inferred segmentation results have been cropped to 513 x 513. So when I use 'EvalRun.m' to evaluate the results, errors occurred in 'MyVOCevalseg.m'.

"Results image '2007_000033' is the wrong size, was 513 x 513, should be 366 x 500."

In this case, should I save the cropped groundtruth label images(513x513) in order to evaluate the results? Or is there any other solutions (such as saving inferred segmentation results to original size--366x500 with 'MatWrite' layer)?

Thank you very much for your time.
Kind regards,

Ruihao