https://www.modelscope.cn/models/iic/cv_ddcolor_image-colorization/summary#ddcolor-%E5%9B%BE%E5%83%8F%E4%B8%8A%E8%89%B2%E6%A8%A1%E5%9E%8B

本博客的代码：https://github.com/xxddccaa/DDColor-webui

下载模型

pip install modelscope


modelscope download --model 'iic/cv_ddcolor_image-colorization' --local_dir './DDColormodel'

推理

pytroch环境：

conda install pytorch torchvision torchaudio pytorch-cuda=12.1 -c pytorch -c nvidia -y

pip install addict modelscope datasets sympy==1.13.1 simplejson sortedcontainers timm

推理：

python
import cv2
from modelscope.outputs import OutputKeys
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks

img_colorization = pipeline(Tasks.image_colorization, 
                       model='./DDColormodel')
img_path = 'https://modelscope.oss-cn-beijing.aliyuncs.com/test/images/audrey_hepburn.jpg'
result = img_colorization(img_path)
cv2.imwrite('result.png', result[OutputKeys.OUTPUT_IMG])

原图：

docker环境

docker run --gpus all --shm-size=32g -it --net host -v ./:/ddcolor kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-pix2pix bash

pip install addict modelscope datasets sympy==1.13.1 simplejson sortedcontainers timm opencv-python

apt-get update
apt-get install -y libgl1-mesa-glx libjpeg-dev libpng-dev libtiff-dev libopencv-dev

docker commit 731cbf160933 kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-ddcolor

docker push kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-ddcolor

webui部署

cd /ssd/xiedong/image_color

docker run --gpus device=2 \
--shm-size=32g \
-it \
--net host \
-v ./ddcolor_app.py:/ddcolor/ddcolor_app.py \
-v ./DDColormodel:/DDColormodel/ \
kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-ddcolor bash

pip install gradio

python /ddcolor/ddcolor_app.py --model_path /DDColormodel --port 7861

用这个镜像也ok：

kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-ddcolor-webui

指标测试

bash
docker run --gpus device=2 \
--shm-size=32g \
-it \
--net host \
-v /ssd/xiedong/image_color:/ssd/xiedong/image_color \
kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-ddcolor-webui bash


pip install evaluation_requirements.txt

转换测试图片：

bash
python ddcolor_inference.py \
--model_path /ssd/xiedong/image_color/DDColormodel/ \
--src_dir /ssd/xiedong/image_color/pytorch-CycleGAN-and-pix2pix/results/tongyong_l2ab_4/testA_35/images \
--dst_dir /ssd/xiedong/image_color/ddcolor_test

得到指标：

python evaluate_colorization.py --results_dir /ssd/xiedong/image_color/ddcolor_test --output_dir /ssd/xiedong/image_color/ddcolor_test_metric  --use_fid

Metric	Mean	Std Dev	Min	Max	Performance Assessment
SSIM	0.887	0.078	0.435	0.983	✅ ​​Excellent​​ (Ideal >0.85)
PSNR (dB)	21.65	3.63	8.91	31.73	⚠️ ​​Good​​ (Typical 20-30dB)
MSE	663.11	860.43	43.67	8349.47	⚠️ ​​Moderate​​ (Lower better)
MAE	16.25	8.20	2.55	72.02	⚠️ ​​Acceptable​​
Color Error	11.39	7.55	1.46	54.99	⚠️ ​​Needs Improvement​​
LPIPS	0.166	0.072	0.025	0.472	✅ ​​Good​​ (Closer to 0 best)
FID	36.51	-	-	-	⚠️ ​​Fair​​ (Ideal <30)

镜像：

kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-ddcolor-webui-metric

webui快速启动

webui 7861

bash
cd /your/path/to/image_color

docker run --gpus device=2 \
--shm-size=32g \
-it \
--net host \
-v ./ddcolor_app.py:/ddcolor/ddcolor_app.py \
-v ./DDColormodel:/DDColormodel/ \
kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-ddcolor-webui-metric bash


python /ddcolor/ddcolor_app.py --model_path /DDColormodel --port 7861

webui 7862

bash
cd /your/path/to/image_color

docker run --gpus device=2 \
--shm-size=32g \
-it \
--net host \
-v ./ddcolor_app.py:/ddcolor/ddcolor_app.py \
-v ./DDColormodel:/DDColormodel/ \
kevinchina/deeplearning:2.5.1-cuda12.1-cudnn9-devel-ddcolor-webui-metric bash


python /ddcolor/ddcolor_app.py --model_path /DDColormodel --port 7862