desnownet_context-aware_deep_network_for_snow_removal's Introduction

DesnowNet: Context-Aware DeepNetwork for Snow Removal

This project is an unofficial code of the paper DesnowNet: Context-Aware DeepNetwork for Snow Removal.

Example

(a) Snowy Image $x$ (b) Estimated snow-free output $\hat{y}$ (c) Estimated snow-free output $y'$ (d) Estimated snow mask $\hat{z}$

Environment

To generate the recovered result you need:

Python3
CPU or NVIDIA GPU + CUDA CuDNN

The required packages can refer to requirements.txt.

Model Checkpoints

create a log directory to save your checkpoints, e.g
```
mkdir ./log
```

How to use the code?

Train

  python3 train.py --device [Device] -r [Root path for training set] -dir [Checkpoints directory] -iter [Iterations] --save_schedule [Save Checkpoints]

Test

  python3 Test.py --device [Device] -dir [Checkpoints directory] -root [Root path for test set] -path [Image Path] --checkpoint [The checkpoint you choose]

Inference

  python3 inference.py --device [Device] -dir [Checkpoints directory] -path [Image Path] --checkpoint [The checkpoint you choose]

train.py will automatically choose the latest checkpoint and continue the training process.

The Improvements

Two-stage training

Train

python3 multi_stage_train.py --device [Device] -iter [Iterations for training RG] -dir [Checkpoints directory] -r [Root path for training set] --save_schedule [Save Checkpoints] --TR_iterations [Iterations for training TR]

Test, Inference: It's the same with the original model.

Predict the product of z and a

Train

python3 train.py --device [Device] -r [Root path for training set] -dir [Checkpoints directory] -iter [Iterations] --save_schedule [Save Checkpoints] --mode za

Test

  python3 test.py --device [Device] -dir [Checkpoints directory]  -root [Root path for test set] -path [Image Path] --checkpoint [The checkpoint you choose] --mode za

Inference:

  python3 inference.py --device [Device] -dir [Checkpoints directory] -path [Image Path] --checkpoint [The checkpoint you choose] --mode za

Reference Code

Pretrained models for Pytorch https://github.com/Cadene/pretrained-models.pytorch/blob/master/pretrainedmodels/models/inceptionv4.py

pytorch-msssim https://github.com/jorge-pessoa/pytorch-msssim

desnownet_context-aware_deep_network_for_snow_removal's People

Contributors

Stargazers

Watchers

desnownet_context-aware_deep_network_for_snow_removal's Issues

Some concerns about the test.

Hi,
Thanks for sharing the code of your great work! Here, we wonder about the detailed setting of testing on Snow 100K dataset. (1)Is this test process implemented on the original resolution or 64 \times 64 or 256 \times 256, or something else? (2) is the calculation of PSNR and SSIM performed on RGB or Y channel?
Looking forward to your answer!
Best,
josh

pre-trained models

Hi, thank you for sharing this reop.
the link you shared for pre-trained models doesn't work. can you provide a new link to the pre-trained models?

@linYDTHU

Recommend Projects

linydthu / desnownet_context-aware_deep_network_for_snow_removal Goto Github PK