MetricsLab

文件目录

• metrics：所有的指标与指标相关的loss，所有的输入都是$$(B,C,H,W)$$形状的张量
  • __init__：用于集合所有的指标与loss，方便导入
  • ssim：（VIFB，Structural similarly-based）Structural Similarity index measure
  • rmse：（VIFB，Structural similarly-based）Root mean squared error
  • ce：（VIFB，Information theory-based）Cross entropy
  • en：（VIFB，Information theory-based）Entropy
  • mi：（VIFB，Information theory-based）Mutural information
  • psnr：（VIFB，Information theory-based）Peak signal-to-noise ration
  • ag：（VIFB，Image feature-based）Average gradient
  • ei：（VIFB，Image feature-based）Edge intensity
  • sd：（VIFB，Image feature-based）Standard deviation
  • sf：（VIFB，Image feature-based）Spatial frequency
  • $$Q^{AB/F}$$：（VIFB，Image feature-based）Gradient-based fusion performance
  • $$Q_{CB}$$：（VIFB，Human perception inspired）Chen-Blum metric
  • $$Q_{CV}$$：（VIFB，Human perception inspired）Chen-Varsheny metric
• imgs：所有的图片
  • RoadScene、TNO：数据集
    • ir：红外图片
    • vis：可见光图片
    • fuse：融合图片
      • U2Fusion等融合方式

Demo

1. 读入图片

   from utils import *
   
   # 通过路径直接导入
   ir_tensor = read_grey_tensor('./imgs/TNO/ir/1.bmp',requires_grad=False)
   vis_tensor = read_grey_tensor('./imgs/TNO/vis/1.bmp',requires_grad=False)
   fuse_tensor = read_grey_tensor('./imgs/TNO/fuse/U2Fusion/1.bmp',requires_grad=True)
   
   # 通过信息间接导入
   ir_tensor = read_grey_tensor(dataset='TNO',category='ir',name='1.bmp',requires_grad=False)
   vis_tensor = read_grey_tensor(dataset='TNO',category='vis',name='1.bmp',requires_grad=False)
   fuse_tensor = read_grey_tensor(dataset='TNO',category='fuse',name='1.bmp',model='U2Fusion',requires_grad=True)