README.md 5.43 KB
Newer Older
mashun1's avatar
diffbir  
mashun1 committed
1
# DiffBIR
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
2

mashun1's avatar
diffbir  
mashun1 committed
3
## 论文
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
4

mashun1's avatar
diffbir  
mashun1 committed
5
**DiffBIR: Towards Blind Image Restoration with Generative Diffusion Prior**
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
6

mashun1's avatar
diffbir  
mashun1 committed
7
* https://arxiv.org/abs/2308.15070
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
8

mashun1's avatar
diffbir  
mashun1 committed
9
## 模型结构
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
10

mashun1's avatar
diffbir  
mashun1 committed
11
第一阶段模型使用8个Swin Transformer blocks(RSTB),每个RSTB中包含6个Swin Transformer Layers(STL),其中head数为6,window size为8.
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
12

mashun1's avatar
diffbir  
mashun1 committed
13
第二阶段模型基于Stable Diffusioin 2.1-base,创建了一个与Unet中encoder block与middle block相同的网络。
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
14

mashun1's avatar
diffbir  
mashun1 committed
15
![Alt text](images/image.png)
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
16

mashun1's avatar
diffbir  
mashun1 committed
17
## 算法原理
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
18

mashun1's avatar
diffbir  
mashun1 committed
19
用途:该算法为两阶段算法,可以提升图像的分辨率。
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
20

mashun1's avatar
diffbir  
mashun1 committed
21
第一阶段使用复原模块,从具有未知和复杂降质的低质量(LQ)图像中恢复清晰图像;
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
22

mashun1's avatar
mashun1 committed
23
24
![Alt text](images/s1.png)

mashun1's avatar
diffbir  
mashun1 committed
25
第二阶段使用生成模块来重新生成丢失的信息。
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
26

mashun1's avatar
mashun1 committed
27
28
![Alt text](images/s2.png)

mashun1's avatar
diffbir  
mashun1 committed
29
## 环境配置
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
30

mashun1's avatar
diffbir  
mashun1 committed
31
### Docker(方法一)
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
32

mashun1's avatar
mashun1 committed
33
34
35
36
37
38
    docker pull image.sourcefind.cn:5000/dcu/admin/base/pytorch:1.13.1-centos7.6-dtk-23.04.1-py39-latest
    docker run --shm-size 10g --network=host --name=diffbir --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined -v 项目地址(绝对路径):/home/ -it <your IMAGE ID> bash
    pip install -r requirements.txt

### Docker(方法二)

mashun1's avatar
diffbir  
mashun1 committed
39
40
41
42
    # 需要在对应的目录下
    docker build -t <IMAGE_NAME>:<TAG> .
    # <your IMAGE ID>用以上拉取的docker的镜像ID替换
    docker run -it --shm-size 10g --network=host --name=diffbir --privileged --device=/dev/kfd --device=/dev/dri --group-add video --cap-add=SYS_PTRACE --security-opt seccomp=unconfined <your IMAGE ID> bash
mashun1's avatar
mashun1 committed
43
   
mashun1's avatar
mashun1 committed
44
45


mashun1's avatar
diffbir  
mashun1 committed
46
47
48
### Anaconda (方法三)
1、关于本项目DCU显卡所需的特殊深度学习库可从光合开发者社区下载安装:
https://developer.hpccube.com/tool/
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
49

mashun1's avatar
diffbir  
mashun1 committed
50
51
52
53
54
55
56
    DTK驱动:dtk23.04.1
    python:python3.9
    torch:1.13.1
    torchvision:0.14.1
    torchaudio:0.13.1
    deepspeed:0.9.2
    apex:0.1
57

mashun1's avatar
diffbir  
mashun1 committed
58
Tips:以上dtk驱动、python、torch等DCU相关工具版本需要严格一一对应
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
59

mashun1's avatar
diffbir  
mashun1 committed
60
2、其它非特殊库参照requirements.txt安装
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
61

mashun1's avatar
diffbir  
mashun1 committed
62
    pip install -r requirements.txt
63

0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
64

mashun1's avatar
diffbir  
mashun1 committed
65
## 数据集
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
66

mashun1's avatar
diffbir  
mashun1 committed
67
下载地址(训练+测试集):https://www.image-net.org/ (imagenet1k)
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
68

mashun1's avatar
diffbir  
mashun1 committed
69
70
71
72
    datasets
        |- train
            |- n01440764
                |- xxx.JPEG
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
73

mashun1's avatar
mashun1 committed
74
75
76
77
78
79
80
下载地址(训练+测试集):https://www.robots.ox.ac.uk/~vgg/data/pets/

    datasets
        |- images
            |- xxx.JPG

注意:以上两个数据集都可以使用,可任选其一。
ziyannchen's avatar
ziyannchen committed
81

mashun1's avatar
diffbir  
mashun1 committed
82
## 训练
83

mashun1's avatar
diffbir  
mashun1 committed
84
### 阶段一
85

mashun1's avatar
diffbir  
mashun1 committed
86
87
88
89
90
91
1、数据准备:该操作用于生成训练以及验证数据路径列表

    python scripts/make_file_list.py \  
    --img_folder [hq_dir_path] \         # 包含图片的文件夹
    --val_size [validation_set_size] \   # 验证集大小
    --save_folder [save_dir_path] \      # 路径列表保存文件夹
ziyannchen's avatar
ziyannchen committed
92
    --follow_links
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
93

mashun1's avatar
diffbir  
mashun1 committed
94
95
96
2、修改配置文件

    修改 `configs/dataset`中相应的yaml配置文件
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
97

mashun1's avatar
diffbir  
mashun1 committed
98
    修改 `configs/train_swinir.yaml`配置文件
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
99

mashun1's avatar
mashun1 committed
100
注意:在修改`configs/dataset`中的配置文件时,可以选择修改"realesrgan"或"codeformer"中的任意一个。
mashun1's avatar
mashun1 committed
101

mashun1's avatar
diffbir  
mashun1 committed
102
103
104
3、训练

    python train.py --config [training_config_path]
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
105

mashun1's avatar
diffbir  
mashun1 committed
106
注意:该阶段训练得到的模型将用于第二阶段的训练。
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
107

mashun1's avatar
diffbir  
mashun1 committed
108
### 阶段二
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
109

mashun1's avatar
diffbir  
mashun1 committed
110
1、模型准备(Stable Diffusion v2.1): https://huggingface.co/stabilityai/stable-diffusion-2-1-base/resolve/main/v2-1_512-ema-pruned.ckpt
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
111
112


mashun1's avatar
diffbir  
mashun1 committed
113
114
115
116
117
118
119
120
121
122
123
124
125
2、初始化模型参数

    python scripts/make_stage2_init_weight.py \
    --cldm_config configs/model/cldm.yaml \
    --sd_weight [sd_v2.1_ckpt_path] \
    --swinir_weight [swinir_ckpt_path] \  # 第一阶段训练得到的模型
    --output [init_weight_output_path]    # 初始化模型保存地址

3、修改配置文件

    修改`configs/train_cldm.yaml`配置文件

4、训练
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
126
127
128

    python train.py --config [training_config_path]

mashun1's avatar
diffbir  
mashun1 committed
129
## 推理
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
130

mashun1's avatar
diffbir  
mashun1 committed
131
### general Image
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
132

mashun1's avatar
diffbir  
mashun1 committed
133
134
模型下载地址:
* https://huggingface.co/lxq007/DiffBIR/resolve/main/general_full_v1.ckpt
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
135

mashun1's avatar
diffbir  
mashun1 committed
136
* https://huggingface.co/lxq007/DiffBIR/resolve/main/general_swinir_v1.ckpt
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
137

mashun1's avatar
diffbir  
mashun1 committed
138
139
140
141
142
143
144
145
146
147
        python inference.py \
        --input inputs/demo/general \
        --config configs/model/cldm.yaml \
        --ckpt weights/general_full_v1.ckpt \  
        --reload_swinir --swinir_ckpt weights/general_swinir_v1.ckpt \  
        --steps 50 \
        --sr_scale 4 \
        --color_fix_type wavelet \
        --output results/demo/general \
        --device cuda [--tiled --tile_size 512 --tile_stride 256]
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
148

mashun1's avatar
diffbir  
mashun1 committed
149
注意:方括号中的参数为可选项,模型也可以替换为在训练阶段得到的
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
150

mashun1's avatar
diffbir  
mashun1 committed
151
### Face Image
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
152

mashun1's avatar
diffbir  
mashun1 committed
153
模型下载地址:
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
154

mashun1's avatar
diffbir  
mashun1 committed
155
* https://huggingface.co/lxq007/DiffBIR/resolve/main/face_full_v1.ckpt
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
156

mashun1's avatar
diffbir  
mashun1 committed
157
for aligned face inputs
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
158

mashun1's avatar
diffbir  
mashun1 committed
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
    python inference_face.py \
    --input inputs/demo/face/aligned \
    --sr_scale 1 \
    --output results/demo/face/aligned \
    --has_aligned \
    --device cuda

for unaligned face inputs

    python inference_face.py \
    --input inputs/demo/face/whole_img \
    --sr_scale 2 \
    --output results/demo/face/whole_img \
    --bg_upsampler DiffBIR \
    --device cuda

## result

恢复后的图像

![Alt text](images/samples_step-004900_e-000008_b-001203.png)

低质量图像

![Alt text](images/lq_step-004900_e-000008_b-001203.png)

### 精度



## 应用场景
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
190

mashun1's avatar
diffbir  
mashun1 committed
191
### 算法类别
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
192

mashun1's avatar
diffbir  
mashun1 committed
193
`图像超分`
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
194

mashun1's avatar
diffbir  
mashun1 committed
195
### 热点应用行业
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
196

mashun1's avatar
mashun1 committed
197
`媒体,科研,教育`
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
198

mashun1's avatar
diffbir  
mashun1 committed
199
## 源码仓库及问题反馈
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
200

mashun1's avatar
mashun1 committed
201
https://developer.hpccube.com/codes/modelzoo/diffbir_pytorch
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
202

mashun1's avatar
mashun1 committed
203
## 参考资料
0x3f3f3f3fun's avatar
0x3f3f3f3fun committed
204

mashun1's avatar
diffbir  
mashun1 committed
205
* https://github.com/XPixelGroup/DiffBIR
mashun1's avatar
mashun1 committed
206
* https://github.com/XPixelGroup/DiffBIR/issues/55