Giters

sithu31296 / semantic-segmentation

SOTA Semantic Segmentation Models in PyTorch

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

About mIoU result on Cityscapes validation dataset

tanveer6715 opened this issue · comments

commented

Hi,

Thanks for your work and providing code. I am just confuse about the mIoU reported in the paper on cityscapes validation dataset which is around 77%. Kindly let me know is this mIoU achieved on pretrained model (imagenet)? I am training DDR-Net-Slim23 from scratch and I am getting around 55% mIoU on validation data of Cityscapes.

commented

Hello, this repo does not replicate all the configurations used in the DDRNet. The training parameters, optimizers, losses, augmentations, etc. might be different from original implementation. To get a result, you need to use larger resolution, batch size, etc. They got 77.8 mIoU with 1024x2048 image size (with multi-scale + flip) I think.

commented

Thank you. I guess they also use validation dataset in training images thats may be the reason of high mIoU.

commented

I believed they didn't use the validation data as the dataset is evaluating on validation set. Try to use configurations as close as possible and see if the results improve.

I will close this issue for now.