[CAM]Other Methods

Author

SEOYEON CHOI

Published

October 18, 2023

https://pythonrepo.com/repo/jacobgil-pytorch-grad-cam

https://github.com/jacobgil/pytorch-grad-cam

Method What it does
GradCAM Weight the 2D activations by the average gradient
GradCAM++ Like GradCAM but uses second order gradients
XGradCAM Like GradCAM but scale the gradients by the normalized activations
AblationCAM Zero out activations and measure how the output drops (this repository includes a fast batched implementation)
ScoreCAM Perbutate the image by the scaled activations and measure how the output drops
EigenCAM Takes the first principle component of the 2D Activations (no class discrimination, but seems to give great results)
EigenGradCAM Like EigenCAM but with class discrimination: First principle component of Activations*Grad. Looks like GradCAM, but cleaner
LayerCAM Spatially weight the activations by positive gradients. Works better especially in lower layers

Import

from pytorch_grad_cam import GradCAM, HiResCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, FullGrad, EigenGradCAM, LayerCAM
import torchvision
import torch
import cv2
import matplotlib.pyplot as plt
from PIL import Image
from torchvision.models import resnet50
from fastai.vision.all import *
from torchvision import transforms
def label_func(f):
    if f[0].isupper():
        return 'cat' 
    else: 
        return 'dog'

Data

path=Path('original_pet') 
files=get_image_files(path)
dls=ImageDataLoaders.from_name_func(path,files,label_func,item_tfms=Resize(512)) 
path_r=Path('random_pet_one')   #랜덤박스넣은사진
files_r=get_image_files(path_r)
dls_r=ImageDataLoaders.from_name_func(path_r,files_r,label_func,item_tfms=Resize(512))

CAM

lrnr=cnn_learner(dls,resnet34,metrics=error_rate)
lrnr.fine_tune(1)
/home/csy/anaconda3/envs/temp_csy/lib/python3.8/site-packages/fastai/vision/learner.py:288: UserWarning: `cnn_learner` has been renamed to `vision_learner` -- please update your code
  warn("`cnn_learner` has been renamed to `vision_learner` -- please update your code")
/home/csy/anaconda3/envs/temp_csy/lib/python3.8/site-packages/torchvision/models/_utils.py:208: UserWarning: The parameter 'pretrained' is deprecated since 0.13 and may be removed in the future, please use 'weights' instead.
  warnings.warn(
/home/csy/anaconda3/envs/temp_csy/lib/python3.8/site-packages/torchvision/models/_utils.py:223: UserWarning: Arguments other than a weight enum or `None` for 'weights' are deprecated since 0.13 and may be removed in the future. The current behavior is equivalent to passing `weights=ResNet34_Weights.IMAGENET1K_V1`. You can also use `weights=ResNet34_Weights.DEFAULT` to get the most up-to-date weights.
  warnings.warn(msg)
epoch train_loss valid_loss error_rate time
0 0.175802 0.017629 0.006089 00:35
epoch train_loss valid_loss error_rate time
0 0.044077 0.004809 0.002030 00:43 
net1=lrnr.model[0]
net2=lrnr.model[1]
net2 = torch.nn.Sequential(
    torch.nn.AdaptiveAvgPool2d(output_size=1), 
    torch.nn.Flatten(),
    torch.nn.Linear(512,out_features=2,bias=False))
net=torch.nn.Sequential(net1,net2)
lrnr2=Learner(dls,net,metrics=accuracy) 
lrnr2.fine_tune(5)
epoch train_loss valid_loss accuracy time
0 0.247475 0.831850 0.725304 00:43
epoch train_loss valid_loss accuracy time
0 0.122386 0.438715 0.914750 00:43
1 0.114546 0.220883 0.916779 00:43
2 0.071775 0.079565 0.963464 00:43
3 0.032215 0.058077 0.977673 00:43
4 0.011436 0.050615 0.982409 00:43 
lrnr_r=cnn_learner(dls_r,resnet34,metrics=error_rate)
lrnr_r.fine_tune(1)
epoch train_loss valid_loss error_rate time
0 0.117688 0.012863 0.005413 00:33
epoch train_loss valid_loss error_rate time
0 0.000532 0.000008 0.000000 00:43 
net1_r=lrnr_r.model[0]
net2_r=lrnr_r.model[1]
net2_r = torch.nn.Sequential(
    torch.nn.AdaptiveAvgPool2d(output_size=1), 
    torch.nn.Flatten(),
    torch.nn.Linear(512,out_features=2,bias=False))
net_r=torch.nn.Sequential(net1_r,net2_r)
lrnr2_r=Learner(dls_r,net_r,metrics=accuracy) 
lrnr2_r.fine_tune(5)
epoch train_loss valid_loss accuracy time
0 0.008425 0.000942 1.000000 00:43
epoch train_loss valid_loss accuracy time
0 0.000006 0.000001 1.000000 00:43
1 0.000001 0.000000 1.000000 00:43
2 0.000001 0.000000 1.000000 00:43
3 0.000000 0.000000 1.000000 00:43
4 0.000000 0.000000 1.000000 00:43 
x_cat, = first(dls.test_dl([PILImage.create(get_image_files(path)[2])]))
x_cat = x_cat.to('cpu')
x_cat_r, = first(dls_r.test_dl([PILImage.create(get_image_files(path_r)[2])]))
x_cat_r = x_cat_r.to('cpu')
x_dog, = first(dls.test_dl([PILImage.create(get_image_files(path)[12])]))
x_dog = x_dog.to('cpu')
x_dog_r, = first(dls_r.test_dl([PILImage.create(get_image_files(path_r)[12])]))
x_dog_r = x_dog_r.to('cpu')

GradCAM

Cat_GradCAM

Cat_GradCAM_Original

gradcam_original = GradCAM(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_gradcam_original = gradcam_original(input_tensor=x_cat,targets=None)
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_gradcam_original.squeeze(), alpha=0.7)

Cat_GradCAM_Randombox

gradcam_randombox = GradCAM(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_gradcam_randombox = gradcam_randombox(input_tensor=x_cat_r,targets=None)
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_gradcam_randombox.squeeze(), alpha=0.7)

Dog_GradCAM

Dog_GradCAM_Original

cam_dog_gradcam_original = gradcam_original(input_tensor=x_dog,targets=None)
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_gradcam_original.squeeze(), alpha=0.7)

Dog_GradCAM_Randombox

cam_dog_gradcam_randombox = gradcam_randombox(input_tensor=x_dog_r,targets=None)
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_gradcam_randombox.squeeze(), alpha=0.7)

HiResCAM

Cat_HiResCAM

Cat_HiResCAM_Original

hirescam_original = HiResCAM(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_hirescam_original = hirescam_original(input_tensor=x_cat,targets=None)
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_hirescam_original.squeeze(), alpha=0.7)

Cat_HiResCAM_Randombox

hirescam_randombox = HiResCAM(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_hirescam_randombox = hirescam_randombox(input_tensor=x_cat_r,targets=None)
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_hirescam_randombox.squeeze(), alpha=0.7)

Dog_HiResCAM

Dog_HiResCAM_Original

cam_dog_hirescam_original = hirescam_original(input_tensor=x_dog,targets=None)
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_hirescam_original.squeeze(), alpha=0.7)

Dog_HiResCAM_Random

cam_dog_hirescam_randombox = hirescam_randombox(input_tensor=x_dog_r,targets=None)
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_hirescam_randombox.squeeze(), alpha=0.7)

ScoreCAM

Cat_ScoreCAM

Cat_ScoreCAM_Original

scorecam_original = ScoreCAM(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_scorecam_original = scorecam_original(input_tensor=x_cat,targets=None)
100%|██████████| 32/32 [00:26<00:00,  1.23it/s]
100%|██████████| 32/32 [00:25<00:00,  1.27it/s]
100%|██████████| 32/32 [00:25<00:00,  1.26it/s]
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_scorecam_original.squeeze(), alpha=0.7)

Cat_ScoreCAM_Randombox

scorecam_randombox = ScoreCAM(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_scorecam_randombox = scorecam_randombox(input_tensor=x_cat_r,targets=None)
100%|██████████| 32/32 [00:25<00:00,  1.25it/s]
100%|██████████| 32/32 [00:25<00:00,  1.26it/s]
100%|██████████| 32/32 [00:25<00:00,  1.26it/s]
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_scorecam_randombox.squeeze(), alpha=0.7)

Dog_ScoreCAM

Dog_ScoreCAM_Original

cam_dog_scorecam_original = scorecam_original(input_tensor=x_dog,targets=None)
100%|██████████| 32/32 [00:25<00:00,  1.25it/s]
100%|██████████| 32/32 [00:25<00:00,  1.27it/s]
100%|██████████| 32/32 [00:24<00:00,  1.30it/s]
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_scorecam_original.squeeze(), alpha=0.7)

Dog_ScoreCAM_Randombox

cam_dog_scorecam_randombox = scorecam_randombox(input_tensor=x_dog_r,targets=None)
100%|██████████| 32/32 [00:24<00:00,  1.30it/s]
100%|██████████| 32/32 [00:24<00:00,  1.30it/s]
100%|██████████| 32/32 [00:24<00:00,  1.29it/s]
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_scorecam_randombox.squeeze(), alpha=0.7)

GradCAMPlusPlus

Cat_GradCAMPlusPlus

Cat_GradCAMPlusPlus_Original

gradcamplusplus_original = GradCAMPlusPlus(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_gradcamplusplus_original = gradcamplusplus_original(input_tensor=x_cat,targets=None)
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_gradcamplusplus_original.squeeze(), alpha=0.7)

Cat_GradCAMPlusPlus_Randombox

gradcamplusplus_randombox = GradCAMPlusPlus(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_gradcamplusplus_randombox = gradcamplusplus_randombox(input_tensor=x_cat_r,targets=None)
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_gradcamplusplus_randombox.squeeze(), alpha=0.7)

Dog_GradCAMPlusPlus

Dog_GradCAMPlusPlus_Original

cam_dog_gradcamplusplus_original = gradcamplusplus_original(input_tensor=x_dog,targets=None)
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_gradcamplusplus_original.squeeze(), alpha=0.7)

Dog_GradCAMPlusPlus_Randombox

cam_dog_gradcamplusplus_randombox = gradcamplusplus_randombox(input_tensor=x_dog_r,targets=None)
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_gradcamplusplus_randombox.squeeze(), alpha=0.7)

AblationCAM

Cat_AblationCAM

Cat_AblationCAM_Original

ablationcam_original = AblationCAM(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_ablationcam_original = ablationcam_original(input_tensor=x_cat,targets=None)
100%|██████████| 16/16 [00:28<00:00,  1.75s/it]
100%|██████████| 16/16 [00:28<00:00,  1.78s/it]
100%|██████████| 16/16 [00:27<00:00,  1.72s/it]
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_ablationcam_original.squeeze(), alpha=0.7)

Cat_AblationCAM_Randombox

ablationcam_randombox = AblationCAM(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_ablationcam_randombox = ablationcam_randombox(input_tensor=x_cat_r,targets=None)
100%|██████████| 16/16 [00:26<00:00,  1.63s/it]
100%|██████████| 16/16 [00:25<00:00,  1.60s/it]
100%|██████████| 16/16 [00:25<00:00,  1.60s/it]
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_ablationcam_randombox.squeeze(), alpha=0.7)

Dog_AblationCAM

Dog_AblationCAM_Original

cam_dog_ablationcam_original = ablationcam_original(input_tensor=x_dog,targets=None)
100%|██████████| 16/16 [00:25<00:00,  1.61s/it]
100%|██████████| 16/16 [00:25<00:00,  1.58s/it]
100%|██████████| 16/16 [00:25<00:00,  1.58s/it]
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_ablationcam_original.squeeze(), alpha=0.7)

Dog_AblationCAM_Randombox

cam_dog_ablationcam_randombox = ablationcam_randombox(input_tensor=x_dog_r,targets=None)
100%|██████████| 16/16 [00:25<00:00,  1.60s/it]
100%|██████████| 16/16 [00:25<00:00,  1.60s/it]
100%|██████████| 16/16 [00:25<00:00,  1.59s/it]
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_ablationcam_randombox.squeeze(), alpha=0.7)

XGradCAM

Cat_XGradCAM

Cat_XGradCAM_Original

xgradcam_original = XGradCAM(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_xgradcam_original = xgradcam_original(input_tensor=x_cat,targets=None)
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_xgradcam_original.squeeze(), alpha=0.7)

Cat_XGradCAM_Randombox

xgradcam_randombox = XGradCAM(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_xgradcam_randombox = xgradcam_randombox(input_tensor=x_cat_r,targets=None)
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_xgradcam_randombox.squeeze(), alpha=0.7)

Dog_XGradCAM

Dog_XGradCAM_Original

cam_dog_xgradcam_original = xgradcam_original(input_tensor=x_dog,targets=None)
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_xgradcam_original.squeeze(), alpha=0.7)

Dog_XGradCAM_Randombox

cam_dog_xgradcam_randombox = xgradcam_randombox(input_tensor=x_dog_r,targets=None)
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_xgradcam_randombox.squeeze(), alpha=0.7)

EigenCAM

Cat_EigenCAM

Cat_EigenCAM_Original

eigencam_original = EigenCAM(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_eigencam_original = eigencam_original(input_tensor=x_cat,targets=None)
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_eigencam_original.squeeze(), alpha=0.7)

Cat_EigenCAM_Randombox

eigencam_randombox = EigenCAM(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_eigencam_randombox = eigencam_randombox(input_tensor=x_cat_r,targets=None)
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_eigencam_randombox.squeeze(), alpha=0.7)

Dog_EigenCAM

Dog_EigenCAM_Original

cam_dog_eigencam_original = eigencam_original(input_tensor=x_dog,targets=None)
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_eigencam_original.squeeze(), alpha=0.7)

Dog_EigenCAM_Randombox

cam_dog_eigencam_randombox = eigencam_randombox(input_tensor=x_dog_r,targets=None)
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_eigencam_randombox.squeeze(), alpha=0.7)

FullGrad

Cat_FullGrad

Cat_FullGrad_Original

fullgrad_original = FullGrad(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
Warning: target_layers is ignored in FullGrad. All bias layers will be used instead
cam_cat_fullgrad_original = fullgrad_original(input_tensor=x_cat,targets=None)
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_fullgrad_original.squeeze(), alpha=0.7)

Cat_FullGrad_Randombox

fullgrad_randombox = FullGrad(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
Warning: target_layers is ignored in FullGrad. All bias layers will be used instead
cam_cat_fullgrad_randombox = fullgrad_randombox(input_tensor=x_cat_r,targets=None)
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_fullgrad_randombox.squeeze(), alpha=0.7)

Dog_FullGrad

Dog_FullGrad_Original

cam_dog_fullgrad_original = fullgrad_original(input_tensor=x_dog,targets=None)
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_fullgrad_original.squeeze(), alpha=0.7)

Dog_FullGrad_Randombox

cam_dog_fullgrad_randombox = fullgrad_randombox(input_tensor=x_dog_r,targets=None)
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_fullgrad_randombox.squeeze(), alpha=0.7)

EigenGradCAM

Cat_EigenGradCAM

Cat_EigenGradCAM_Original

eigengradcam_original = EigenGradCAM(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_eigengradcam_original = eigengradcam_original(input_tensor=x_cat,targets=None)
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_eigengradcam_original.squeeze(), alpha=0.7)

Cat_EigenGradCAM_Randombox

eigengradcam_randombox = EigenGradCAM(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_eigengradcam_randombox = eigengradcam_randombox(input_tensor=x_cat_r,targets=None)
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_eigengradcam_randombox.squeeze(), alpha=0.7)

Dog_EigenGradCAM

Dog_EigenGradCAM_Original

cam_dog_eigengradcam_original = eigengradcam_original(input_tensor=x_dog,targets=None)
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_eigengradcam_original.squeeze(), alpha=0.7)

Dog_EigenGradCAM_Randombox

cam_dog_eigengradcam_randombox = eigengradcam_randombox(input_tensor=x_dog_r,targets=None)
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_eigengradcam_randombox.squeeze(), alpha=0.7)

LayerCAM

Cat_LayerCAM

Cat_LayerCAM_Original

layercam_original = LayerCAM(model=lrnr2.model.to('cpu'), target_layers=lrnr2.model[0][-1])
cam_cat_layercam_original = layercam_original(input_tensor=x_cat,targets=None)
dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(cam_cat_layercam_original.squeeze(), alpha=0.7)

Cat_LayerCAM_Randombox

layercam_randombox = LayerCAM(model=lrnr2_r.model.to('cpu'), target_layers=lrnr2_r.model[0][-1])
cam_cat_layercam_randombox = layercam_randombox(input_tensor=x_cat_r,targets=None)
dls.train.decode((x_cat_r,))[0].squeeze().show()
plt.imshow(cam_cat_layercam_randombox.squeeze(), alpha=0.7)

Dog_LayerCAM

Dog_LayerCAM_Original

cam_dog_layercam_original = layercam_original(input_tensor=x_dog,targets=None)
dls.train.decode((x_dog,))[0].squeeze().show()
plt.imshow(cam_dog_layercam_original.squeeze(), alpha=0.7)

Dog_LayerCAM_Randombox

cam_dog_layercam_randombox = layercam_randombox(input_tensor=x_dog_r,targets=None)
dls.train.decode((x_dog_r,))[0].squeeze().show()
plt.imshow(cam_dog_layercam_randombox.squeeze(), alpha=0.7)

Figure

Figure_Original

fig, ((ax1,ax2,ax3,ax4),
      (ax5,ax6,ax7,ax8),
      (ax9,ax10,ax11,ax12),
      (ax13,ax14,ax15,ax16),
     (ax17,ax18,ax19,ax20)) = plt.subplots(5,4) 
plt.title('Original')
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax1)
ax1.imshow(cam_cat_gradcam_original.squeeze(), alpha=0.7)
ax1.set_title("GradCAM CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_dog_gradcam_original.squeeze(), alpha=0.7)
ax2.set_title("GradCAM DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax3)
ax3.imshow(cam_cat_hirescam_original.squeeze(), alpha=0.7)
ax3.set_title("HiResCAM CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax4)
ax4.imshow(cam_dog_hirescam_original.squeeze(), alpha=0.7)
ax4.set_title("HiResCAM DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax5)
ax5.imshow(cam_cat_scorecam_original.squeeze(), alpha=0.7)
ax5.set_title("ScoreCAM CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax6)
ax6.imshow(cam_dog_scorecam_original.squeeze(), alpha=0.7)
ax6.set_title("ScoreCAM DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax7)
ax7.imshow(cam_cat_gradcamplusplus_original.squeeze(), alpha=0.7)
ax7.set_title("GradCAMPlusPlus CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax8)
ax8.imshow(cam_dog_gradcamplusplus_original.squeeze(), alpha=0.7)
ax8.set_title("GradCAMPlusPlus DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax9)
ax9.imshow(cam_cat_ablationcam_original.squeeze(), alpha=0.7)
ax9.set_title("AblationCAM CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax10)
ax10.imshow(cam_dog_ablationcam_original.squeeze(), alpha=0.7)
ax10.set_title("AblationCAM DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax11)
ax11.imshow(cam_cat_xgradcam_original.squeeze(), alpha=0.7)
ax11.set_title("XGradCAM CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax12)
ax12.imshow(cam_dog_xgradcam_original.squeeze(), alpha=0.7)
ax12.set_title("XGradCAM DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax13)
ax13.imshow(cam_cat_eigencam_original.squeeze(), alpha=0.7)
ax13.set_title("EigenCAM CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax14)
ax14.imshow(cam_dog_eigencam_original.squeeze(), alpha=0.7)
ax14.set_title("EigenCAM DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax15)
ax15.imshow(cam_cat_fullgrad_original.squeeze(), alpha=0.7)
ax15.set_title("FullGrad CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax16)
ax16.imshow(cam_dog_fullgrad_original.squeeze(), alpha=0.7)
ax16.set_title("FullGrad DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax17)
ax17.imshow(cam_cat_eigengradcam_original.squeeze(), alpha=0.7)
ax17.set_title("EigenGradCAM CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax18)
ax18.imshow(cam_dog_eigengradcam_original.squeeze(), alpha=0.7)
ax18.set_title("EigenGradCAM DOG PART")
#
dls.train.decode((x_cat,))[0].squeeze().show(ax=ax19)
ax19.imshow(cam_cat_layercam_original.squeeze(), alpha=0.7)
ax19.set_title("LayerCAM CAT PART")
#
dls.train.decode((x_dog,))[0].squeeze().show(ax=ax20)
ax20.imshow(cam_dog_layercam_original.squeeze(), alpha=0.7)
ax20.set_title("LayerCAM DOG PART")
#
fig.set_figwidth(20)            
fig.set_figheight(20)
fig.tight_layout()

Figure_Randombox

fig, ((ax1,ax2,ax3,ax4),
      (ax5,ax6,ax7,ax8),
      (ax9,ax10,ax11,ax12),
      (ax13,ax14,ax15,ax16),
     (ax17,ax18,ax19,ax20)) = plt.subplots(5,4) 
plt.title('Randombox')
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax1)
ax1.imshow(cam_cat_gradcam_randombox.squeeze(), alpha=0.7)
ax1.set_title("GradCAM CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_dog_gradcam_randombox.squeeze(), alpha=0.7)
ax2.set_title("GradCAM DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax3)
ax3.imshow(cam_cat_hirescam_randombox.squeeze(), alpha=0.7)
ax3.set_title("HiResCAM CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax4)
ax4.imshow(cam_dog_hirescam_randombox.squeeze(), alpha=0.7)
ax4.set_title("HiResCAM DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax5)
ax5.imshow(cam_cat_scorecam_randombox.squeeze(), alpha=0.7)
ax5.set_title("ScoreCAM CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax6)
ax6.imshow(cam_dog_scorecam_randombox.squeeze(), alpha=0.7)
ax6.set_title("ScoreCAM DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax7)
ax7.imshow(cam_cat_gradcamplusplus_randombox.squeeze(), alpha=0.7)
ax7.set_title("GradCAMPlusPlus CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax8)
ax8.imshow(cam_dog_gradcamplusplus_randombox.squeeze(), alpha=0.7)
ax8.set_title("GradCAMPlusPlus DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax9)
ax9.imshow(cam_cat_ablationcam_randombox.squeeze(), alpha=0.7)
ax9.set_title("AblationCAM CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax10)
ax10.imshow(cam_dog_ablationcam_randombox.squeeze(), alpha=0.7)
ax10.set_title("AblationCAM DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax11)
ax11.imshow(cam_cat_xgradcam_randombox.squeeze(), alpha=0.7)
ax11.set_title("XGradCAM CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax12)
ax12.imshow(cam_dog_xgradcam_randombox.squeeze(), alpha=0.7)
ax12.set_title("XGradCAM DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax13)
ax13.imshow(cam_cat_eigencam_randombox.squeeze(), alpha=0.7)
ax13.set_title("EigenCAM CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax14)
ax14.imshow(cam_dog_eigencam_randombox.squeeze(), alpha=0.7)
ax14.set_title("EigenCAM DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax15)
ax15.imshow(cam_cat_fullgrad_randombox.squeeze(), alpha=0.7)
ax15.set_title("FullGrad CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax16)
ax16.imshow(cam_dog_fullgrad_randombox.squeeze(), alpha=0.7)
ax16.set_title("FullGrad DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax17)
ax17.imshow(cam_cat_eigengradcam_randombox.squeeze(), alpha=0.7)
ax17.set_title("EigenGradCAM CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax18)
ax18.imshow(cam_dog_eigengradcam_randombox.squeeze(), alpha=0.7)
ax18.set_title("EigenGradCAM DOG PART")
#
dls.train.decode((x_cat_r,))[0].squeeze().show(ax=ax19)
ax19.imshow(cam_cat_layercam_randombox.squeeze(), alpha=0.7)
ax19.set_title("LayerCAM CAT PART")
#
dls.train.decode((x_dog_r,))[0].squeeze().show(ax=ax20)
ax20.imshow(cam_dog_layercam_randombox.squeeze(), alpha=0.7)
ax20.set_title("LayerCAM DOG PART")
#
fig.set_figwidth(20)            
fig.set_figheight(20)
fig.tight_layout()