Is this result correct or not?

Question

Is this result correct or not?

ly-jq opened this issue a year ago · comments

After training the deeplab3plus resnet101 model with 30000 iters, the mean accuracy achieves to 0.678442 and miou 0.593765. This result is lower for your conclusion in the Cityscapes dataset, it is correct or not. I didn't use pretrained weight to train.
deeplab3p.log

Gongfan Fang · Answer 1 · Mon Mar 27 2023 15:44:57 GMT+0800 (China Standard Time)

Hi @ly-jq, could you please provide the training configs? The deeplabv3plus_resnet101 was contributed by a community user.

haidrbilal · Answer 2 · Thu Jun 22 2023 16:22:31 GMT+0800 (China Standard Time)

According to encoder decoder atrous separable conv paper, they achieved 82.1% test set accuracy. What hyperparameters did they use while training?

DC-SegNet · Answer 3 · Wed Aug 02 2023 09:33:22 GMT+0800 (China Standard Time)

Hi @ly-jq, could you please provide the training configs? The deeplabv3plus_resnet101 was contributed by a community user.

Has the problem been solved?

wmx2000219 · Answer 4 · Sat Jan 20 2024 22:34:21 GMT+0800 (China Standard Time)

Hi @ly-jq, could you please provide the training configs? The deeplabv3plus_resnet101 was contributed by a community user.
Hi, I recently trained deeplabv3+Xception on the VOC2012 dataset, and after 150 epochs the miou metric on the validation set was only 0.26, and the acc metric had 0.72. I don't know what's wrong with the code, is it because of the small number of epochs trained?