halbielee / EPS

Official PyTorch implementation of "Railroad is not a Train: Saliency as Pseudo-pixel Supervision for Weakly Supervised Semantic Segmentation", CVPR2021
89 stars 13 forks source link

COCO training codes #6

Closed pigcv closed 3 years ago

pigcv commented 3 years ago

Hi~ Could you provide COCO training codes, to help us reproduce the results on COCO?

Thanks!

halbielee commented 3 years ago

Hello, @pigcv again!

I will provide the training code & dataset for COCO soon!

pigcv commented 3 years ago

Hello, @pigcv again!

I will provide the training code & dataset for COCO soon!

Thanks for you reply. Another question is that why deeplab v1 performs better than deeplab v2? Which Deeplab V1 repo are you using? Thanks again!

halbielee commented 3 years ago

We basically use the network in here.

But it only provides DeepLab-V2. Since the backbone networks are the same between DeepLabV1 & V2, we manually make DeepLabV1 from DeepLabV2. We use the same pre-trained model!

halbielee commented 3 years ago

Hello, @pigcv

I uploaded the training data for MS-COCO 2014 (segmentation mask for train/validation set) and the training code. You can check them with new script files.

Also, I updated the checkpoints, training logs, performances for each method and each dataset.

Please see README.md!

pigcv commented 3 years ago

Hello, @pigcv

I uploaded the training data for MS-COCO 2014 (segmentation mask for train/validation set) and the training code. You can check them with new script files.

Also, I updated the checkpoints, training logs, performances for each method and each dataset.

Please see README.md!

Thanks very much!