YonghaoXu

Hi, the attention module could bring about 1% and 2% improvement on the mIoU metric for the ``GTA-5->Cityscapes`` and ``Synthia->Cityscapes`` scenarios, respectively.

Hi, Yawei. Thank you for your interest in our work. Following the original mean-teacher model, the exponential moving average method is adopted to implement this mechenism. Since the teacher network...

This question also puzzles me for a period. I can't explain it very scientifically, but I am willing to share my intuitive guess. Usually, an ensemble model of multiple base...

Hi @lauraset , thanks for your interest in this work. As discussed with Yawei, although there is no specific design to guarantee that the teacher network would yield more accurate...

Hi @lauraset , you are welcome. If Gaussian noise is too strong, you may also consider adding dropout and other image transformations to conduct stochastic augmentation, as introduced in the...

Hi @lauraset , your understanding is correct. If there are batch normalization or dropout layers in the teacher network, we need to switch between `teacher_net.train()` and `teacher_net.eval()` for training and...

Hi, thank you for sharing these reproduction results. As mentioned in the ``Empirical Observations`` in the ``README`` doc, the obtained mIoU results may fluctuate a little bit due to the...

Hi, I have removed the self-ensembling and attention modules in the code and got a similar result as you did. The detailed accuracy records are: ``GTA2Cityscapes_epoch1batch500tgt_miu_232`` ``GTA2Cityscapes_epoch1batch1000tgt_miu_242`` ``GTA2Cityscapes_epoch1batch1500tgt_miu_280`` ``GTA2Cityscapes_epoch1batch2000tgt_miu_293`` ``GTA2Cityscapes_epoch1batch4000tgt_miu_293``...

Hi, the URL of the pretrained model has been updated. Feel free to download it.

Hi @sandeep-25, we follow the previous literature [1] to set the ` train_num_array`. You can also change the value in ` train_num_array` with other settings to check the performance with...