search

Haochen-Wang409 / U2PL

[CVPR'22 & IJCV'24] Semi-Supervised Semantic Segmentation Using Unreliable Pseudo-Labels & Using Unreliable Pseudo-Labels for Label-Efficient Semantic Segmentation
Apache License 2.0
436 stars 61 forks source link

Questions about Augmented PASCAL-VOC2012 #95

Closed DeepHM closed 2 years ago

DeepHM commented 2 years ago

As with the use of the Augmented PASCAL-VOC2012 dataset in other semi-supervised semantic segmentation studies (i.e., CPS, PS_MT) , I would like to reproduce your study in 1/8 (1323) of the Augmented PASCAL-VOC. However, it seems to me that this github is training code from Full(1464). Any code that trains on 1/8 (1323) of Augmented PASCAL-VOC? Or is it possible by simply changing the options in the 'config.yaml' file?

Haochen-Wang409 commented 2 years ago

Just change the data_list and n_sup in config.yaml.

DeepHM commented 2 years ago

Thank you !

DeepHM commented 2 years ago

Hello. PASCAL_VOC does not contain "eval.sh". How can I do this?

Haochen-Wang409 commented 2 years ago

There is no need to run eval.sh for PASCAL-VOC experments. The performance has been reported during training.

While for Cityscapes, we apply slide window evaluation follow previous SOTA AEL, and thus eval.sh is necessary.

DeepHM commented 2 years ago

Thank you !!

DeepHM commented 2 years ago

Hello. Can I know the gpu information (gpu name(type), number of gpu, and memory usage of gpu) and training time in your experiments?

If possible, I would like to know roughly about VOC and Cicyscapes, respectively. (especially training time and gpu memory usage)

Thank you !

Haochen-Wang409 commented 2 years ago

Please refer to Note section of README.

DeepHM commented 2 years ago

oh.. thank you!

DeepHM commented 2 years ago

Hello. In your paper, the study only appears to have the model trained with Resnet101.

If you have a model trained with Resnet50 on PASCAL VOC2012, can you please share? (model file with Resnet50 in PASCAL VOC2012)

Haochen-Wang409 commented 2 years ago

We did not train R50 for any datasets.

DeepHM commented 2 years ago

Thanks for your kind reply.