# 實驗紀錄 ###### tags: `論文進度` ## 實驗環境 | 作業系統 | Windows 10 21H2 | | -------- | -------- | | CPU | Intel i7-9750H | |GPU|NVIDIA GeForce RTX2070| |記憶體|16 GB * 2 / DDR4 / 2667 MHz| ### <font color="#f00">debug</font> 將 Pytorch 降版至與CUDA相容最低版 1.4.0,解決 CUDA error 錯誤 ``` RuntimeError: CUDA error: unknown error ``` ### GPU 以目前實驗結果看來,目前的評估指標與論文結果都滿相近的,觀察結果看來,在 GoPro、REDS dataset 中甚至有更好的 SSIM 值 ## Baseline **1. Denoise - SIDD dataset**  | 方法 | PSNR | | -------- | -------- | | HINet (論文) | 39.82 | |HINet (ours) |39.8231| **2. Deblur - GoPro dataset**  | 方法 | PSNR |SSIM| | -------- | -------- | -------- | | HINet (論文) | 32.77 |0.959| |HINet (ours) |32.7712|0.9950| **3. Deblur - REDS dataset**  | 方法 | PSNR |SSIM| | -------- | -------- | -------- | | HINet (論文) | 28.83|0.862| |HINet (ours) |28.8267|0.9801| **4. Derain - Rain13k dataset**  * For Test100 : | 方法 | PSNR |SSIM| | -------- | -------- | -------- | | HINet (論文) | 30.29 |0.906| |HINet (ours) |30.2864|0.9063| * For Rain100H :  | 方法 | PSNR |SSIM| | -------- | -------- | -------- | | HINet (論文) | 30.65 |0.894| |HINet (ours) |30.6495|0.8944| * For Rain100L :  | 方法 | PSNR |SSIM| | -------- | -------- | -------- | | HINet (論文) | 37.28 |0.970| |HINet (ours) |37.2767|0.9698| * For Test2800 :  | 方法 | PSNR |SSIM| | -------- | -------- | -------- | | HINet (論文) | 33.91 |0.941| |HINet (ours) |33.9079|0.9412| * For Test1200 :  | 方法 | PSNR |SSIM| | -------- | -------- | -------- | | HINet (論文) | 33.05 |0.919| |HINet (ours) |33.0473|0.9194| ## Dockerfile ```shell= FROM meadml/cuda10.1-cudnn7-devel-ubuntu18.04-python3.6 ## Install some basic utilities RUN apt-get update RUN apt-get install -y \ git \ zip \ sudo \ libx11-6 \ build-essential \ ca-certificates \ wget \ curl \ tmux \ htop \ nano \ vim ## RUN apt update RUN apt-get install ffmpeg libsm6 libxext6 -y ## RUN pip install addict \ future \ lmdb \ numpy \ opencv-python \ Pillow \ pyyaml \ requests \ scikit-image \ scipy \ tb-nightly \ tqdm \ yapf RUN pip install torch===1.4.0+cu92 torchvision===0.5.0+cu92 -f https://download.pytorch.org/whl/torch_stable.html ```
×
Sign in
Email
Password
Forgot password
or
By clicking below, you agree to our
terms of service
.
Sign in via Facebook
Sign in via Twitter
Sign in via GitHub
Sign in via Dropbox
Sign in with Wallet
Wallet (
)
Connect another wallet
New to HackMD?
Sign up