HantingChen comments

Results 107 comments of


                                            HantingChen

Some question about the mindspore code different from paper

The num_layers should be changed to 12. Note that the mindspore code is a 4 layers demo.

How to get Single Image Inference (output) for Deraining and Denoising

For derain: python main.py --dir_data $DATA_PATH --pretrain $MODEL_PATH --scale 1 --derain So if you want to get the deraining results, you may set the arg: --derain instead of --derain_test 1

How to get Single Image Inference (output) for Deraining and Denoising

You can modify : self.derain_dataroot = os.path.join(args.dir_data, "RainTrainL") self.derain_img_list = search(self.derain_dataroot, "rainstreak") self.derain_test = os.path.join(args.dir_data, "Rain100L") self.derain_lr_test = search(self.derain_test, "rain") self.derain_hr_test = [path.replace("rainy/","no") for path in self.derain_lr_test] in [Lines](https://github.com/huawei-noah/Pretrained-IPT/blob/fd4b1c7d5a853df778ea83ddd92271bd301a5f98/data/srdata.py#L49-L53) to:...

NameError: name 'model' is not defined

Thanks for the feedback. We have fixed the bug.

PSNR很小

可能是文件路径或图片错误，或者读取的预训练模型错误。

test question

You can add "DICM" in https://github.com/huawei-noah/Pretrained-IPT/blob/32ce841c18ca33c31fa15907ac21bbddd1583519/data/__init__.py#L40, and then put the data in /Pretrained-IPT-main/test_data/benchmark/DICM .

How to fine-tune a new dataset?

We only release inference code in github. You can try the code in Mindspore, which includes the training code: https://gitee.com/mindspore/mindspore/tree/master/model_zoo/research/cv/IPT

重新训练时，图像尺寸与代码提供的尺寸不一致，该如何调整相关超参数？

你好，可以采用 --patch_size 256 --patch_dim 8 --n_feats 9进行超参数调整

请问degeneration方法的代码在哪里呢？

可以参考https://gitee.com/mindspore/models/blob/master/research/cv/IPT/src/data/srdata.py

train

You can refer to https://gitee.com/mindspore/models/tree/master/research/cv/IPT. Thanks!