[CAM]Dataset1

Author

SEOYEON CHOI

Published

January 5, 2024

https://docs.fast.ai/data.external.html

Import

from pytorch_grad_cam import GradCAM, HiResCAM, ScoreCAM, GradCAMPlusPlus, AblationCAM, XGradCAM, EigenCAM, FullGrad, EigenGradCAM, LayerCAM
from pytorch_grad_cam.utils.model_targets import ClassifierOutputTarget
import torch 
from fastai.vision.all import *
from fastai.vision import *
import cv2
import numpy as np
import os
os.environ['CUDA_LAUNCH_BLOCKING'] = "1"
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
import cv2
import numpy as np
import matplotlib.pyplot as plt
from PIL import ImageDraw
from PIL import ImageFont
from PIL import ImageFile
from PIL import Image
ImageFile.LOAD_TRUNCATED_IMAGES = True
from torchvision import *
import os
import rpy2
import rpy2.robjects as ro 
from rpy2.robjects.vectors import FloatVector 
from rpy2.robjects.packages import importr

Data

path = untar_data(URLs.CALTECH_101);path
Path('/home/csy/.fastai/data/caltech_101')
path.ls()
(#102) [Path('/home/csy/.fastai/data/caltech_101/gerenuk'),Path('/home/csy/.fastai/data/caltech_101/ceiling_fan'),Path('/home/csy/.fastai/data/caltech_101/electric_guitar'),Path('/home/csy/.fastai/data/caltech_101/schooner'),Path('/home/csy/.fastai/data/caltech_101/sunflower'),Path('/home/csy/.fastai/data/caltech_101/brain'),Path('/home/csy/.fastai/data/caltech_101/dolphin'),Path('/home/csy/.fastai/data/caltech_101/nautilus'),Path('/home/csy/.fastai/data/caltech_101/panda'),Path('/home/csy/.fastai/data/caltech_101/chair')...]
dls = ImageDataLoaders.from_folder(path, valid_pct=0.2, seed=42, item_tfms=Resize(512), 
                                   batch_tfms=[*aug_transforms(), Normalize.from_stats(*imagenet_stats)])
dls.show_batch()

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 2.786667 1.086679 0.280635 00:46
epoch train_loss valid_loss error_rate time
0 0.953599 0.546696 0.141685 00:59 
net1=lrnr.model[0]
net2=lrnr.model[1] 
net2
Sequential(
  (0): AdaptiveConcatPool2d(
    (ap): AdaptiveAvgPool2d(output_size=1)
    (mp): AdaptiveMaxPool2d(output_size=1)
  )
  (1): fastai.layers.Flatten(full=False)
  (2): BatchNorm1d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (3): Dropout(p=0.25, inplace=False)
  (4): Linear(in_features=1024, out_features=512, bias=False)
  (5): ReLU(inplace=True)
  (6): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (7): Dropout(p=0.5, inplace=False)
  (8): Linear(in_features=512, out_features=102, bias=False)
)
net2 = torch.nn.Sequential(
    torch.nn.AdaptiveAvgPool2d(output_size=1), 
    torch.nn.Flatten(),
    torch.nn.Linear(512,out_features=102,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 2.200760 3.150540 0.371444 00:59
epoch train_loss valid_loss accuracy time
0 1.145681 1.464551 0.653173 00:59
1 1.119536 1.774413 0.589716 00:58
2 0.751984 0.828347 0.771335 00:59
3 0.415494 0.496882 0.862144 00:59
4 0.232705 0.418902 0.882385 00:59

Step

x, = first(dls.test_dl([PILImage.create(get_image_files(path)[20])]))
dls.train.decode((x,))[0].squeeze().show()

x = x.to('cpu')
camimg = torch.einsum('ij,jkl -> ikl', net2[2].weight, net1(x).squeeze())
ebayesthresh = importr('EbayesThresh').ebayesthresh

power_threshed=np.array(ebayesthresh(FloatVector(torch.tensor(camimg[0].detach().reshape(-1))**2)))
ybar_threshed = np.where(power_threshed>80,torch.tensor(camimg[0].detach().reshape(-1)).cpu(),0)
ybar_threshed = torch.tensor(ybar_threshed.reshape(16,16))

power_threshed2=np.array(ebayesthresh(FloatVector(torch.tensor(camimg[1].detach().reshape(-1))**2)))
ybar_threshed2 = np.where(power_threshed2>80,torch.tensor(camimg[1].detach().reshape(-1)).cpu(),0)
ybar_threshed2 = torch.tensor(ybar_threshed2.reshape(16,16))
UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed=np.array(ebayesthresh(FloatVector(torch.tensor(camimg[0].detach().reshape(-1))**2)))
<ipython-input-870-11d406c6ed63>:4: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed = np.where(power_threshed>80,torch.tensor(camimg[0].detach().reshape(-1)).cpu(),0)
<ipython-input-870-11d406c6ed63>:7: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed2=np.array(ebayesthresh(FloatVector(torch.tensor(camimg[1].detach().reshape(-1))**2)))
<ipython-input-870-11d406c6ed63>:8: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed2 = np.where(power_threshed2>80,torch.tensor(camimg[1].detach().reshape(-1)).cpu(),0)
fig, (ax1,ax2,ax3) = plt.subplots(1,3) 
# 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.set_title("Input image")
# 
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow((ybar_threshed).to("cpu").detach(),alpha=0.5,extent=(0,48,48,0),interpolation='bilinear',cmap='cool')
# ax2.set_title("CAT PART")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax3)
ax3.imshow((ybar_threshed2).to("cpu").detach(),alpha=0.5,extent=(0,48,48,0),interpolation='bilinear',cmap='cool')
# ax3.set_title("DOG PART")
#
fig.set_figwidth(12)            
fig.set_figheight(12)
fig.tight_layout()

  • 판단 근거가 강할 수록 파란색 -> 보라색

mode 1

# test=camimg_o[0]-torch.min(camimg_o[0])
A1=torch.exp(-0.05*(ybar_threshed))
A2 = 1 - A1
fig, (ax1,ax2) = plt.subplots(1,2) 
# 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(A2.data.to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("MODE1 WEIGHTT")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(A1.data.to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("MODE1 RES WEIGHT")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

# mode 1 res
X1=np.array(A1.to("cpu").detach(),dtype=np.float32)
Y1=torch.Tensor(cv2.resize(X1,(512,512),interpolation=cv2.INTER_LINEAR))
x1=x.squeeze().to('cpu')*Y1-torch.min(x.squeeze().to('cpu'))*Y1

# mode 1
X12=np.array(A2.to("cpu").detach(),dtype=np.float32)
Y12=torch.Tensor(cv2.resize(X12,(512,512),interpolation=cv2.INTER_LINEAR))
x12=x.squeeze().to('cpu')*Y12-torch.min(x.squeeze().to('cpu'))*Y12

- 1st CAM 분리

fig, (ax1) = plt.subplots(1,1) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.set_title("ORIGINAL")
fig.set_figwidth(4)            
fig.set_figheight(4)
fig.tight_layout()
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x12*0.3).squeeze().show(ax=ax1)  #MODE1
(x1*0.1).squeeze().show(ax=ax2)  #MODE1_res
ax1.set_title("MODE1")
ax2.set_title("MODE1 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

x1 = x1.reshape(1,3,512,512)
net1.to('cpu')
net2.to('cpu')
Sequential(
  (0): AdaptiveAvgPool2d(output_size=1)
  (1): Flatten(start_dim=1, end_dim=-1)
  (2): Linear(in_features=512, out_features=102, bias=False)
)
camimg1 = torch.einsum('ij,jkl -> ikl', net2[2].weight, net1(x1).squeeze())
power_threshed3=np.array(ebayesthresh(FloatVector(torch.tensor(camimg1[0].detach().reshape(-1))**2)))
ybar_threshed3 = np.where(power_threshed3>10,torch.tensor(camimg1[0].detach().reshape(-1)),0)
ybar_threshed3 = torch.tensor(ybar_threshed3.reshape(16,16))

power_threshed4=np.array(ebayesthresh(FloatVector(torch.tensor(camimg1[1].detach().reshape(-1))**2)))
ybar_threshed4 = np.where(power_threshed4>10,torch.tensor(camimg1[1].detach().reshape(-1)),0)
ybar_threshed4 = torch.tensor(ybar_threshed4.reshape(16,16))
UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed3=np.array(ebayesthresh(FloatVector(torch.tensor(camimg1[0].detach().reshape(-1))**2)))
<ipython-input-900-292f842a7fbc>:2: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed3 = np.where(power_threshed3>10,torch.tensor(camimg1[0].detach().reshape(-1)),0)
<ipython-input-900-292f842a7fbc>:5: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed4=np.array(ebayesthresh(FloatVector(torch.tensor(camimg1[1].detach().reshape(-1))**2)))
<ipython-input-900-292f842a7fbc>:6: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed4 = np.where(power_threshed4>10,torch.tensor(camimg1[1].detach().reshape(-1)),0)

- mode1 res

fig, (ax1,ax2) = plt.subplots(1,2) 
# 
(x1*0.25).squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed3,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("CAT PART")
#
(x1*0.25).squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed4,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DOG PART")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

- 첫번째 CAM 결과와 비교

fig, (ax1,ax2) = plt.subplots(1,2) 
# 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("1ST CAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed3,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("2ND CAM")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

- 2nd CAM 분리

# test1=camimg1[0]-torch.min(camimg1[0])
A3 = torch.exp(-0.05*(ybar_threshed3))
A4 = 1 - A3
fig, (ax1,ax2) = plt.subplots(1,2) 
# 
x1.squeeze().show(ax=ax2)
dls.train.decode((x1,))[0].squeeze().show(ax=ax1)
ax1.imshow(A3.data.to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("MODE2 RES WEIGHT")
#
x1.squeeze().show(ax=ax2)
dls.train.decode((x1,))[0].squeeze().show(ax=ax2)
ax2.imshow(A4.data.to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("MODE2 WEIGHT")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

X2=np.array(A3.to("cpu").detach(),dtype=np.float32)
Y2=torch.Tensor(cv2.resize(X2,(512,512),interpolation=cv2.INTER_LINEAR))
x2=(x1)*Y2-torch.min((x1)*Y2)

X22=np.array(A4.to("cpu").detach(),dtype=np.float32)
Y22=torch.Tensor(cv2.resize(X22,(512,512),interpolation=cv2.INTER_LINEAR))
x22=(x1)*Y22-torch.min((x1)*Y22)
fig, (ax1) = plt.subplots(1,1) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.set_title("ORIGINAL")
fig.set_figwidth(4)            
fig.set_figheight(4)
fig.tight_layout()
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x12*0.3).squeeze().show(ax=ax1)  #MODE1
(x1*0.1).squeeze().show(ax=ax2)  #MODE1_res
ax1.set_title("MODE1")
ax2.set_title("MODE1 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x22+1.7).squeeze().show(ax=ax1)  #MODE2
(x2*0.1).squeeze().show(ax=ax2)  #MODE2_res
ax1.set_title("MODE2")
ax2.set_title("MODE2 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

x2 = x2.reshape(1,3,512,512)
net1.to('cpu')
net2.to('cpu')
Sequential(
  (0): AdaptiveAvgPool2d(output_size=1)
  (1): Flatten(start_dim=1, end_dim=-1)
  (2): Linear(in_features=512, out_features=102, bias=False)
)
camimg2 = torch.einsum('ij,jkl -> ikl', net2[2].weight, net1(x2).squeeze())
power_threshed5=np.array(ebayesthresh(FloatVector(torch.tensor(camimg2[0].detach().reshape(-1))**2)))
ybar_threshed5 = np.where(power_threshed5>40,torch.tensor(camimg2[0].detach().reshape(-1)),0)
ybar_threshed5 = torch.tensor(ybar_threshed5.reshape(16,16))

power_threshed6=np.array(ebayesthresh(FloatVector(torch.tensor(camimg2[1].detach().reshape(-1))**2)))
ybar_threshed6 = np.where(power_threshed6>40,torch.tensor(camimg2[1].detach().reshape(-1)),0)
ybar_threshed6 = torch.tensor(ybar_threshed6.reshape(16,16))
UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed5=np.array(ebayesthresh(FloatVector(torch.tensor(camimg2[0].detach().reshape(-1))**2)))
<ipython-input-919-ba6e482cf5c4>:2: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed5 = np.where(power_threshed5>40,torch.tensor(camimg2[0].detach().reshape(-1)),0)
<ipython-input-919-ba6e482cf5c4>:5: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed6=np.array(ebayesthresh(FloatVector(torch.tensor(camimg2[1].detach().reshape(-1))**2)))
<ipython-input-919-ba6e482cf5c4>:6: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed6 = np.where(power_threshed6>40,torch.tensor(camimg2[1].detach().reshape(-1)),0)

- mode2 res 에 CAM 결과 올리기

fig, (ax1, ax2) = plt.subplots(1,2) 
#
x2.squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed5,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("CAT PART")
#
x2.squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed6,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DOG PART")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

fig, (ax1,ax2,ax3) = plt.subplots(1,3) 
# 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("1ST CAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed3,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("2ND CAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax3)
ax3.imshow(ybar_threshed5,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax3.set_title("3RD CAM")
fig.set_figwidth(12)            
fig.set_figheight(12)
fig.tight_layout()

mode 3 만들기

# test2=camimg2[0]-torch.min(camimg2[0])
A5 = torch.exp(-0.05*(ybar_threshed5))
A6 = 1 - A5
fig, (ax1, ax2) = plt.subplots(1,2) 
#
x2.squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed5,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("CAT PART")
#
x2.squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed6,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DOG PART")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

#mode 3 res
X3=np.array(A5.to("cpu").detach(),dtype=np.float32)
Y3=torch.Tensor(cv2.resize(X3,(512,512),interpolation=cv2.INTER_LINEAR))
x3=x2*Y3-torch.min(x2*Y3)
# mode 3
X32=np.array(A6.to("cpu").detach(),dtype=np.float32)
Y32=torch.Tensor(cv2.resize(X32,(512,512),interpolation=cv2.INTER_LINEAR))
x32=x2*Y32-torch.min(x2*Y32)
import pickle
fig, (ax1) = plt.subplots(1,1) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.set_title("ORIGINAL")
fig.set_figwidth(4)            
fig.set_figheight(4)
fig.tight_layout()

with open('CALTECH_101_hcam_cat_plt1.pkl', 'wb') as file:
    pickle.dump(fig, file)
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x12*0.3).squeeze().show(ax=ax1)  #MODE1
(x1*0.1).squeeze().show(ax=ax2)  #MODE1_res
ax1.set_title("MODE1")
ax2.set_title("MODE1 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
with open('CALTECH_101_hcam_cat_plt2.pkl', 'wb') as file:
    pickle.dump(fig, file)
    
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x22+1.7).squeeze().show(ax=ax1)  #MODE2
(x2*0.1).squeeze().show(ax=ax2)  #MODE2_res
ax1.set_title("MODE2")
ax2.set_title("MODE2 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
with open('CALTECH_101_hcam_cat_plt3.pkl', 'wb') as file:
    pickle.dump(fig, file)
    
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x32+2).squeeze().show(ax=ax1)  #MODE3
(x3*0.1).squeeze().show(ax=ax2)  #MODE3_res
ax1.set_title("MODE3")
ax2.set_title("MODE3 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

with open('CALTECH_101_hcam_cat_plt4.pkl', 'wb') as file:
    pickle.dump(fig, file)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

with open('CALTECH_101_hcam_cat_plt1.pkl', 'rb') as file:
    hcam_cat_plt = pickle.load(file)
plt.show(hcam_cat_plt)

with open('CALTECH_101_hcam_cat_plt2.pkl', 'rb') as file:
    hcam_cat_plt = pickle.load(file)
plt.show(hcam_cat_plt)

with open('CALTECH_101_hcam_cat_plt3.pkl', 'rb') as file:
    hcam_cat_plt = pickle.load(file)
plt.show(hcam_cat_plt)

with open('CALTECH_101_hcam_cat_plt4.pkl', 'rb') as file:
    hcam_cat_plt = pickle.load(file)
plt.show(hcam_cat_plt)

fig, (ax1) = plt.subplots(1,1) 
(x12*-0.3).squeeze().show(ax=ax1)  #MODE1
ax1.set_title("MODE1")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

with open('CALTECH_101_hcam_cat_plt_mode1.pkl', 'wb') as file:
    pickle.dump(fig, file)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

with open('CALTECH_101_hcam_cat_plt_mode1.pkl', 'rb') as file:
    hcam_cat_plt_mode1 = pickle.load(file)
plt.show(hcam_cat_plt_mode1)

fig, (ax1) = plt.subplots(1,1) 
(-x12*0.3 + -x22+1.7).squeeze().show(ax=ax1)  #MODE1+MODE2
ax1.set_title("MODE1+MODE2")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

with open('CALTECH_101_hcam_cat_plt_mode2.pkl', 'wb') as file:
    pickle.dump(fig, file)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

with open('CALTECH_101_hcam_cat_plt_mode2.pkl', 'rb') as file:
    hcam_cat_plt_mode2 = pickle.load(file)
plt.show(hcam_cat_plt_mode2)

fig, (ax1) = plt.subplots(1,1) 
(x12*-0.3 + -x22+1.7 + -x32+2).squeeze().show(ax=ax1)  #MODE1+MODE2+MODE3
ax1.set_title("MODE1+MODE2+MODE3")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

with open('CALTECH_101_hcam_cat_plt_mode3.pkl', 'wb') as file:
    pickle.dump(fig, file)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

with open('CALTECH_101_hcam_cat_plt_mode3.pkl', 'rb') as file:
    hcam_cat_plt_mode3 = pickle.load(file)
plt.show(hcam_cat_plt_mode3)

Other Methods

model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
gradcam = GradCAM(model=model, target_layers=target_layer)
/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)
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
hirescam = HiResCAM(model=model, target_layers=target_layer)
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
scorecam = ScoreCAM(model=model, target_layers=target_layer)
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
gradcamplusplus = GradCAMPlusPlus(model=model, target_layers=target_layer)
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
ablationcam = AblationCAM(model=model, target_layers=target_layer)
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
xgradcam = XGradCAM(model=model, target_layers=target_layer)
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
eigencam = EigenCAM(model=model, target_layers=target_layer)
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
fullgrad = FullGrad(model=model, target_layers=target_layer)
Warning: target_layers is ignored in FullGrad. All bias layers will be used instead
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
eigengradcam = EigenGradCAM(model=model, target_layers=target_layer)
model = resnet34(pretrained=True)

target_layer = [model.layer4[-1]]
layercam = LayerCAM(model=model, target_layers=target_layer)

시도

x, = first(dls.test_dl([PILImage.create(get_image_files(path)[20])]))
x = x.to('cpu')
dls.train.decode((x,))[0].squeeze().show()

cam_gradcam = gradcam(input_tensor=x,targets=None)
cam_hirescam = hirescam(input_tensor=x,targets=None)
cam_scorecam = scorecam(input_tensor=x,targets=None)
100%|██████████| 32/32 [00:23<00:00,  1.38it/s]
cam_gradcamplusplus = gradcamplusplus(input_tensor=x,targets=None)
cam_ablationcam = ablationcam(input_tensor=x,targets=None)
100%|██████████| 16/16 [00:26<00:00,  1.63s/it]
cam_xgradcam = xgradcam(input_tensor=x,targets=None)
cam_eigencam = eigencam(input_tensor=x,targets=None)
cam_fullgrad = fullgrad(input_tensor=x,targets=None)
cam_eigengradcam = eigengradcam(input_tensor=x,targets=None)
cam_layercam = layercam(input_tensor=x,targets=None)
fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_gradcam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART GradCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_gradcam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART GradCAM")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_hirescam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART HiResCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_hirescam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART HiResCAM")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_scorecam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART ScoreCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_scorecam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART ScoreCAM")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_gradcamplusplus.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART GradCAMPlusPlus")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_gradcamplusplus.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART GradCAMPlusPlus")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_ablationcam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART AblationCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_ablationcam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART AblationCAM")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_xgradcam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART XGradCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_xgradcam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART XGradCAM")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_eigencam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART EigenCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_eigencam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART EigenCAM")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_fullgrad.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART FullGrad")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_fullgrad.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART FullGrad")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, (ax1,ax2) = plt.subplots(1,2) 
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(-cam_layercam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("NORMAL PART LayerCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_layercam.squeeze(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DISEASE PART LayerCAM")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

fig, ((ax1,ax2,ax3,ax4),
      (ax5,ax6,ax7,ax8),
      (ax9,ax10,ax11,ax12)) = plt.subplots(3,4) 
plt.title('CALTECH 101 Dataset')
#
dls.train.decode((x,))[0].squeeze().show(ax=ax1)
ax1.imshow(cam_gradcam.squeeze(), alpha=0.7)
ax1.set_title("GradCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_hirescam.squeeze(), alpha=0.7)
ax2.set_title("HiResCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax3)
ax3.imshow(cam_scorecam.squeeze(), alpha=0.7)
ax3.set_title("ScoreCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax4)
ax4.imshow(cam_gradcamplusplus.squeeze(), alpha=0.7)
ax4.set_title("GradCAMPlusPlus")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax5)
ax5.imshow(cam_ablationcam.squeeze(), alpha=0.7)
ax5.set_title("AblationCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax6)
ax6.imshow(cam_xgradcam.squeeze(), alpha=0.7)
ax6.set_title("XGradCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax7)
ax7.imshow(cam_eigencam.squeeze(), alpha=0.7)
ax7.set_title("EigenCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax8)
ax8.imshow(cam_fullgrad.squeeze(), alpha=0.7)
ax8.set_title("FullGrad")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax9)
ax9.imshow(cam_eigengradcam.squeeze(), alpha=0.7)
ax9.set_title("EigenGradCAM")
#
dls.train.decode((x,))[0].squeeze().show(ax=ax10)
ax10.imshow(cam_layercam.squeeze(), alpha=0.7)
ax10.set_title("LayerCAM")

(dls.train.decode((x,))[0].squeeze()*0).show(ax=ax11)
(dls.train.decode((x,))[0].squeeze()*0).show(ax=ax12)
#
fig.set_figwidth(20)            
fig.set_figheight(20)
fig.tight_layout()

with open('CALTECH_101_fig_randombox_plt.pkl', 'wb') as file:
    pickle.dump(fig, file)

with open('CALTECH_101_fig_randombox_plt.pkl', 'rb') as file:
    fig_randombox_plt = pickle.load(file)
fig_randombox_plt.show()

randombox Image download

list(path.ls())
[Path('/home/csy/.fastai/data/caltech_101/gerenuk'),
 Path('/home/csy/.fastai/data/caltech_101/ceiling_fan'),
 Path('/home/csy/.fastai/data/caltech_101/electric_guitar'),
 Path('/home/csy/.fastai/data/caltech_101/schooner'),
 Path('/home/csy/.fastai/data/caltech_101/sunflower'),
 Path('/home/csy/.fastai/data/caltech_101/brain'),
 Path('/home/csy/.fastai/data/caltech_101/dolphin'),
 Path('/home/csy/.fastai/data/caltech_101/nautilus'),
 Path('/home/csy/.fastai/data/caltech_101/panda'),
 Path('/home/csy/.fastai/data/caltech_101/chair'),
 Path('/home/csy/.fastai/data/caltech_101/trilobite'),
 Path('/home/csy/.fastai/data/caltech_101/stapler'),
 Path('/home/csy/.fastai/data/caltech_101/beaver'),
 Path('/home/csy/.fastai/data/caltech_101/snoopy'),
 Path('/home/csy/.fastai/data/caltech_101/windsor_chair'),
 Path('/home/csy/.fastai/data/caltech_101/ant'),
 Path('/home/csy/.fastai/data/caltech_101/cougar_face'),
 Path('/home/csy/.fastai/data/caltech_101/octopus'),
 Path('/home/csy/.fastai/data/caltech_101/menorah'),
 Path('/home/csy/.fastai/data/caltech_101/inline_skate'),
 Path('/home/csy/.fastai/data/caltech_101/dollar_bill'),
 Path('/home/csy/.fastai/data/caltech_101/chandelier'),
 Path('/home/csy/.fastai/data/caltech_101/tick'),
 Path('/home/csy/.fastai/data/caltech_101/garfield'),
 Path('/home/csy/.fastai/data/caltech_101/crab'),
 Path('/home/csy/.fastai/data/caltech_101/Faces'),
 Path('/home/csy/.fastai/data/caltech_101/flamingo_head'),
 Path('/home/csy/.fastai/data/caltech_101/car_side'),
 Path('/home/csy/.fastai/data/caltech_101/saxophone'),
 Path('/home/csy/.fastai/data/caltech_101/crocodile_head'),
 Path('/home/csy/.fastai/data/caltech_101/cup'),
 Path('/home/csy/.fastai/data/caltech_101/water_lilly'),
 Path('/home/csy/.fastai/data/caltech_101/starfish'),
 Path('/home/csy/.fastai/data/caltech_101/Motorbikes'),
 Path('/home/csy/.fastai/data/caltech_101/dragonfly'),
 Path('/home/csy/.fastai/data/caltech_101/barrel'),
 Path('/home/csy/.fastai/data/caltech_101/laptop'),
 Path('/home/csy/.fastai/data/caltech_101/flamingo'),
 Path('/home/csy/.fastai/data/caltech_101/wild_cat'),
 Path('/home/csy/.fastai/data/caltech_101/scorpion'),
 Path('/home/csy/.fastai/data/caltech_101/helicopter'),
 Path('/home/csy/.fastai/data/caltech_101/emu'),
 Path('/home/csy/.fastai/data/caltech_101/pyramid'),
 Path('/home/csy/.fastai/data/caltech_101/BACKGROUND_Google'),
 Path('/home/csy/.fastai/data/caltech_101/rooster'),
 Path('/home/csy/.fastai/data/caltech_101/binocular'),
 Path('/home/csy/.fastai/data/caltech_101/llama'),
 Path('/home/csy/.fastai/data/caltech_101/pizza'),
 Path('/home/csy/.fastai/data/caltech_101/wrench'),
 Path('/home/csy/.fastai/data/caltech_101/cellphone'),
 Path('/home/csy/.fastai/data/caltech_101/strawberry'),
 Path('/home/csy/.fastai/data/caltech_101/scissors'),
 Path('/home/csy/.fastai/data/caltech_101/crayfish'),
 Path('/home/csy/.fastai/data/caltech_101/soccer_ball'),
 Path('/home/csy/.fastai/data/caltech_101/wheelchair'),
 Path('/home/csy/.fastai/data/caltech_101/grand_piano'),
 Path('/home/csy/.fastai/data/caltech_101/sea_horse'),
 Path('/home/csy/.fastai/data/caltech_101/lobster'),
 Path('/home/csy/.fastai/data/caltech_101/crocodile'),
 Path('/home/csy/.fastai/data/caltech_101/pigeon'),
 Path('/home/csy/.fastai/data/caltech_101/yin_yang'),
 Path('/home/csy/.fastai/data/caltech_101/umbrella'),
 Path('/home/csy/.fastai/data/caltech_101/lamp'),
 Path('/home/csy/.fastai/data/caltech_101/airplanes'),
 Path('/home/csy/.fastai/data/caltech_101/stop_sign'),
 Path('/home/csy/.fastai/data/caltech_101/kangaroo'),
 Path('/home/csy/.fastai/data/caltech_101/Faces_easy'),
 Path('/home/csy/.fastai/data/caltech_101/lotus'),
 Path('/home/csy/.fastai/data/caltech_101/accordion'),
 Path('/home/csy/.fastai/data/caltech_101/cannon'),
 Path('/home/csy/.fastai/data/caltech_101/watch'),
 Path('/home/csy/.fastai/data/caltech_101/mayfly'),
 Path('/home/csy/.fastai/data/caltech_101/bass'),
 Path('/home/csy/.fastai/data/caltech_101/stegosaurus'),
 Path('/home/csy/.fastai/data/caltech_101/brontosaurus'),
 Path('/home/csy/.fastai/data/caltech_101/euphonium'),
 Path('/home/csy/.fastai/data/caltech_101/elephant'),
 Path('/home/csy/.fastai/data/caltech_101/gramophone'),
 Path('/home/csy/.fastai/data/caltech_101/butterfly'),
 Path('/home/csy/.fastai/data/caltech_101/buddha'),
 Path('/home/csy/.fastai/data/caltech_101/okapi'),
 Path('/home/csy/.fastai/data/caltech_101/Leopards'),
 Path('/home/csy/.fastai/data/caltech_101/ibis'),
 Path('/home/csy/.fastai/data/caltech_101/ewer'),
 Path('/home/csy/.fastai/data/caltech_101/pagoda'),
 Path('/home/csy/.fastai/data/caltech_101/anchor'),
 Path('/home/csy/.fastai/data/caltech_101/camera'),
 Path('/home/csy/.fastai/data/caltech_101/dalmatian'),
 Path('/home/csy/.fastai/data/caltech_101/metronome'),
 Path('/home/csy/.fastai/data/caltech_101/ferry'),
 Path('/home/csy/.fastai/data/caltech_101/headphone'),
 Path('/home/csy/.fastai/data/caltech_101/cougar_body'),
 Path('/home/csy/.fastai/data/caltech_101/hawksbill'),
 Path('/home/csy/.fastai/data/caltech_101/minaret'),
 Path('/home/csy/.fastai/data/caltech_101/mandolin'),
 Path('/home/csy/.fastai/data/caltech_101/bonsai'),
 Path('/home/csy/.fastai/data/caltech_101/rhino'),
 Path('/home/csy/.fastai/data/caltech_101/ketch'),
 Path('/home/csy/.fastai/data/caltech_101/revolver'),
 Path('/home/csy/.fastai/data/caltech_101/joshua_tree'),
 Path('/home/csy/.fastai/data/caltech_101/platypus'),
 Path('/home/csy/.fastai/data/caltech_101/hedgehog')]
len(list(path.ls()))
5394
temp_img = PILImage.create(get_image_files(path)[20])
temp_img = temp_img.resize([512,512], resample=None, box=None, reducing_gap=None)
(temp_w, temp_h) = (temp_img.shape[0], temp_img.shape[1])
temp_a = random.uniform(0, temp_w*0.8)
temp_b = random.uniform(0, temp_h*0.3)
temp_shape = [(temp_a, temp_b), (temp_a+200, temp_b+60)]
temp_font = ImageFont.truetype("DejaVuSans.ttf", round(temp_h*0.04))
temp_name = (str(list(path.ls())[20])+'-'+str(20)).split('/')[-1] + '.jpg'
temp_fname = temp_name.split('.')[0].split('-')[0]
temp_name
'dollar_bill-20.jpg'
temp_fname
'dollar_bill'
temp_img1 = ImageDraw.Draw(temp_img)  
temp_img1.rectangle(temp_shape, fill ="white", outline ="black")
ImageDraw.Draw(temp_img).text((temp_a+5, temp_b+15), temp_fname, (0,0,0), font=temp_font)
plt.imshow(temp_img)
temp_img.save("caltech_101_random/"+temp_name)

get_image_files(path)[9143]
Path('/home/csy/.fastai/data/caltech_101/hedgehog/image_0017.jpg')
len(path.ls())
102
# for i in range(len(str(list(path.ls())))-1) :
#     img = PILImage.create(get_image_files(path)[i])
#     img = img.resize([512,512], resample=None, box=None, reducing_gap=None)
#     (w, h) = (img.shape[0], img.shape[1])
#     a = random.uniform(0, w*0.8)
#     b = random.uniform(0, h*0.3)
#     shape = [(a, b), (a+200, b+60)]
#     font = ImageFont.truetype("DejaVuSans.ttf", round(h*0.04))
#     name = (os.path.dirname(get_image_files(path)[i]).split('/')[-1]+'-'+str(i)).split('/')[-1] + '.jpg'
#     fname = name.split('.')[0].split('-')[0]
#     img1 = ImageDraw.Draw(img)  
#     img1.rectangle(shape, fill ="white", outline ="black")
#     ImageDraw.Draw(img).text((a+5, b+15), fname, (0,0,0), font=font)
#     img.save("caltech_101_random/"+name)

randombox

path_r=Path('caltech_101_random')   #랜덤박스넣은사진
path_r.ls()
(#5394) [Path('caltech_101_random/Motorbikes-2934.jpg'),Path('caltech_101_random/beaver-782.jpg'),Path('caltech_101_random/Motorbikes-2975.jpg'),Path('caltech_101_random/beaver-796.jpg'),Path('caltech_101_random/BACKGROUND_Google-3983.jpg'),Path('caltech_101_random/Motorbikes-2599.jpg'),Path('caltech_101_random/laptop-3308.jpg'),Path('caltech_101_random/airplanes-5324.jpg'),Path('caltech_101_random/airplanes-5329.jpg'),Path('caltech_101_random/flamingo_head-1941.jpg')...]
files_r=get_image_files(path_r)
def label_func(f):
    return f[0].split('-')[0]
dls_r=ImageDataLoaders.from_name_func(path_r,files,label_func,item_tfms=Resize(512))

(2) 학습

lrnr_r=cnn_learner(dls_r,resnet34,metrics=error_rate)
lrnr_r.fine_tune(1)
epoch train_loss valid_loss error_rate time
0 2.194400 0.814425 0.234694 00:30
epoch train_loss valid_loss error_rate time
0 0.633236 0.386290 0.114100 00:36 
net1_r=lrnr_r.model[0]
net2_r=lrnr_r.model[1] 
net2_r
Sequential(
  (0): AdaptiveConcatPool2d(
    (ap): AdaptiveAvgPool2d(output_size=1)
    (mp): AdaptiveMaxPool2d(output_size=1)
  )
  (1): fastai.layers.Flatten(full=False)
  (2): BatchNorm1d(1024, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (3): Dropout(p=0.25, inplace=False)
  (4): Linear(in_features=1024, out_features=512, bias=False)
  (5): ReLU(inplace=True)
  (6): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (7): Dropout(p=0.5, inplace=False)
  (8): Linear(in_features=512, out_features=23, bias=False)
)
net2_r = torch.nn.Sequential(
    torch.nn.AdaptiveAvgPool2d(output_size=1), 
    torch.nn.Flatten(),
    torch.nn.Linear(512,out_features=23,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 0.990693 7.214967 0.175325 00:36
epoch train_loss valid_loss accuracy time
0 0.191813 0.626072 0.789425 00:36
1 0.138365 0.371003 0.892393 00:36
2 0.059181 0.026200 0.996289 00:36
3 0.019740 0.011075 0.997217 00:36
4 0.006568 0.010148 0.997217 00:36

Step by step

x_r, = first(dls.test_dl([PILImage.create(get_image_files(path_r)[532])]))
dls.train.decode((x_r,))[0].squeeze().show()

camimg_r = torch.einsum('ij,jkl -> ikl', net2_r[2].weight, net1_r(x_r).squeeze())
ebayesthresh = importr('EbayesThresh').ebayesthresh

power_threshed=np.array(ebayesthresh(FloatVector(torch.tensor(camimg_r[0].detach().reshape(-1))**2)))
ybar_threshed = np.where(power_threshed>20,torch.tensor(camimg_r[0].detach().reshape(-1)).cpu(),0)
ybar_threshed = torch.tensor(ybar_threshed.reshape(16,16))

power_threshed2=np.array(ebayesthresh(FloatVector(torch.tensor(camimg_r[1].detach().reshape(-1))**2)))
ybar_threshed2 = np.where(power_threshed2>20,torch.tensor(camimg_r[1].detach().reshape(-1)).cpu(),0)
ybar_threshed2 = torch.tensor(ybar_threshed2.reshape(16,16))
UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed=np.array(ebayesthresh(FloatVector(torch.tensor(camimg_r[0].detach().reshape(-1))**2)))
<ipython-input-695-1fb834e23ce9>:4: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed = np.where(power_threshed>20,torch.tensor(camimg_r[0].detach().reshape(-1)).cpu(),0)
<ipython-input-695-1fb834e23ce9>:7: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed2=np.array(ebayesthresh(FloatVector(torch.tensor(camimg_r[1].detach().reshape(-1))**2)))
<ipython-input-695-1fb834e23ce9>:8: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed2 = np.where(power_threshed2>20,torch.tensor(camimg_r[1].detach().reshape(-1)).cpu(),0)
fig, (ax1,ax2,ax3) = plt.subplots(1,3) 
# 
dls.train.decode((x_r,))[0].squeeze().show(ax=ax1)
ax1.set_title("Input image")
# 
dls.train.decode((x_r,))[0].squeeze().show(ax=ax2)
ax2.imshow((ybar_threshed).to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
# ax2.set_title("CAT PART")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax3)
ax3.imshow((ybar_threshed2).to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
# ax3.set_title("DOG PART")
#
fig.set_figwidth(12)            
fig.set_figheight(12)
fig.tight_layout()

  • 판단 근거가 강할 수록 파란색 -> 보라색

mode 1

# test=camimg_o[0]-torch.min(camimg_o[0])
A1=torch.exp(-0.05*(ybar_threshed))
A2 = 1 - A1
fig, (ax1,ax2) = plt.subplots(1,2) 
# 
dls.train.decode((x_r,))[0].squeeze().show(ax=ax1)
ax1.imshow(A2.data.to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("MODE1 WEIGHTT")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax2)
ax2.imshow(A1.data.to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("MODE1 RES WEIGHT")
#
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

# mode 1 res
X1=np.array(A1.to("cpu").detach(),dtype=np.float32)
Y1=torch.Tensor(cv2.resize(X1,(512,512),interpolation=cv2.INTER_LINEAR))
x1=x_r.squeeze().to('cpu')*Y1-torch.min(x_r.squeeze().to('cpu'))*Y1

# mode 1
X12=np.array(A2.to("cpu").detach(),dtype=np.float32)
Y12=torch.Tensor(cv2.resize(X12,(512,512),interpolation=cv2.INTER_LINEAR))
x12=x_r.squeeze().to('cpu')*Y12-torch.min(x_r.squeeze().to('cpu'))*Y12

- 1st CAM 분리

fig, (ax1) = plt.subplots(1,1) 
dls.train.decode((x_r,))[0].squeeze().show(ax=ax1)
ax1.set_title("ORIGINAL")
fig.set_figwidth(4)            
fig.set_figheight(4)
fig.tight_layout()
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(x12*-1.5).squeeze().show(ax=ax1)  #MODE1
(x1*0.5).squeeze().show(ax=ax2)  #MODE1_res
ax1.set_title("MODE1")
ax2.set_title("MODE1 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

x1 = x1.reshape(1,3,512,512)
net1_r.to('cpu')
net2_r.to('cpu')
Sequential(
  (0): AdaptiveAvgPool2d(output_size=1)
  (1): Flatten(start_dim=1, end_dim=-1)
  (2): Linear(in_features=512, out_features=23, bias=False)
)
camimg1_r = torch.einsum('ij,jkl -> ikl', net2_r[2].weight, net1_r(x1).squeeze())
power_threshed3=np.array(ebayesthresh(FloatVector(torch.tensor(camimg1_r[0].detach().reshape(-1))**2)))
ybar_threshed3 = np.where(power_threshed3>20,torch.tensor(camimg1_r[0].detach().reshape(-1)),0)
ybar_threshed3 = torch.tensor(ybar_threshed3.reshape(16,16))

power_threshed4=np.array(ebayesthresh(FloatVector(torch.tensor(camimg1_r[1].detach().reshape(-1))**2)))
ybar_threshed4 = np.where(power_threshed4>20,torch.tensor(camimg1_r[1].detach().reshape(-1)),0)
ybar_threshed4 = torch.tensor(ybar_threshed4.reshape(16,16))
UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed3=np.array(ebayesthresh(FloatVector(torch.tensor(camimg1_r[0].detach().reshape(-1))**2)))
<ipython-input-724-c83bfb4d70dd>:2: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed3 = np.where(power_threshed3>20,torch.tensor(camimg1_r[0].detach().reshape(-1)),0)
<ipython-input-724-c83bfb4d70dd>:5: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed4=np.array(ebayesthresh(FloatVector(torch.tensor(camimg1_r[1].detach().reshape(-1))**2)))
<ipython-input-724-c83bfb4d70dd>:6: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed4 = np.where(power_threshed4>20,torch.tensor(camimg1_r[1].detach().reshape(-1)),0)

- mode1 res

fig, (ax1,ax2) = plt.subplots(1,2) 
# 
(x1*0.25).squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed3,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("CAT PART")
#
(x1*0.25).squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed4,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DOG PART")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

- 첫번째 CAM 결과와 비교

fig, (ax1,ax2) = plt.subplots(1,2) 
# 
dls.train.decode((x_r,))[0].squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("1ST CAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed3,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("2ND CAM")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

- 2nd CAM 분리

# test1=camimg1[0]-torch.min(camimg1[0])
A3 = torch.exp(-0.05*(ybar_threshed3))
A4 = 1 - A3
fig, (ax1,ax2) = plt.subplots(1,2) 
# 
x1.squeeze().show(ax=ax2)
dls.train.decode((x1,))[0].squeeze().show(ax=ax1)
ax1.imshow(A3.data.to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("MODE2 RES WEIGHT")
#
x1.squeeze().show(ax=ax2)
dls.train.decode((x1,))[0].squeeze().show(ax=ax2)
ax2.imshow(A4.data.to("cpu").detach(),alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("MODE2 WEIGHT")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

X2=np.array(A3.to("cpu").detach(),dtype=np.float32)
Y2=torch.Tensor(cv2.resize(X2,(512,512),interpolation=cv2.INTER_LINEAR))
x2=(x1)*Y2-torch.min((x1)*Y2)

X22=np.array(A4.to("cpu").detach(),dtype=np.float32)
Y22=torch.Tensor(cv2.resize(X22,(512,512),interpolation=cv2.INTER_LINEAR))
x22=(x1)*Y22-torch.min((x1)*Y22)
fig, (ax1) = plt.subplots(1,1) 
dls.train.decode((x_r,))[0].squeeze().show(ax=ax1)
ax1.set_title("ORIGINAL")
fig.set_figwidth(4)            
fig.set_figheight(4)
fig.tight_layout()
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(x12*-1.5).squeeze().show(ax=ax1)  #MODE1
(x1*0.5).squeeze().show(ax=ax2)  #MODE1_res
ax1.set_title("MODE1")
ax2.set_title("MODE1 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x22+1).squeeze().show(ax=ax1)  #MODE2
(x2*0.5).squeeze().show(ax=ax2)  #MODE2_res
ax1.set_title("MODE2")
ax2.set_title("MODE2 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

x2 = x2.reshape(1,3,512,512)
net1_r.to('cpu')
net2_r.to('cpu')
Sequential(
  (0): AdaptiveAvgPool2d(output_size=1)
  (1): Flatten(start_dim=1, end_dim=-1)
  (2): Linear(in_features=512, out_features=23, bias=False)
)
camimg2_r = torch.einsum('ij,jkl -> ikl', net2_r[2].weight, net1_r(x2).squeeze())
power_threshed5=np.array(ebayesthresh(FloatVector(torch.tensor(camimg2_r[0].detach().reshape(-1))**2)))
ybar_threshed5 = np.where(power_threshed5>10,torch.tensor(camimg2_r[0].detach().reshape(-1)),0)
ybar_threshed5 = torch.tensor(ybar_threshed5.reshape(16,16))

power_threshed6=np.array(ebayesthresh(FloatVector(torch.tensor(camimg2_r[1].detach().reshape(-1))**2)))
ybar_threshed6 = np.where(power_threshed6>10,torch.tensor(camimg2_r[1].detach().reshape(-1)),0)
ybar_threshed6 = torch.tensor(ybar_threshed6.reshape(16,16))
UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed5=np.array(ebayesthresh(FloatVector(torch.tensor(camimg2_r[0].detach().reshape(-1))**2)))
<ipython-input-746-f3a70d6c47ba>:2: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed5 = np.where(power_threshed5>10,torch.tensor(camimg2_r[0].detach().reshape(-1)),0)
<ipython-input-746-f3a70d6c47ba>:5: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  power_threshed6=np.array(ebayesthresh(FloatVector(torch.tensor(camimg2_r[1].detach().reshape(-1))**2)))
<ipython-input-746-f3a70d6c47ba>:6: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).
  ybar_threshed6 = np.where(power_threshed6>10,torch.tensor(camimg2_r[1].detach().reshape(-1)),0)

- mode2 res 에 CAM 결과 올리기

fig, (ax1, ax2) = plt.subplots(1,2) 
#
x2.squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed5,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("CAT PART")
#
x2.squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed6,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DOG PART")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

fig, (ax1,ax2,ax3) = plt.subplots(1,3) 
# 
dls.train.decode((x_r,))[0].squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("1ST CAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed3,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("2ND CAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax3)
ax3.imshow(ybar_threshed5,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax3.set_title("3RD CAM")
fig.set_figwidth(12)            
fig.set_figheight(12)
fig.tight_layout()

mode 3 만들기

# test2=camimg2[0]-torch.min(camimg2[0])
A5 = torch.exp(-0.05*(ybar_threshed5))
A6 = 1 - A5
fig, (ax1, ax2) = plt.subplots(1,2) 
#
x2.squeeze().show(ax=ax1)
ax1.imshow(ybar_threshed5,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax1.set_title("CAT PART")
#
x2.squeeze().show(ax=ax2)
ax2.imshow(ybar_threshed6,alpha=0.5,extent=(0,511,511,0),interpolation='bilinear',cmap='cool')
ax2.set_title("DOG PART")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

#mode 3 res
X3=np.array(A5.to("cpu").detach(),dtype=np.float32)
Y3=torch.Tensor(cv2.resize(X3,(512,512),interpolation=cv2.INTER_LINEAR))
x3=x2*Y3-torch.min(x2*Y3)
# mode 3
X32=np.array(A6.to("cpu").detach(),dtype=np.float32)
Y32=torch.Tensor(cv2.resize(X32,(512,512),interpolation=cv2.INTER_LINEAR))
x32=x2*Y32-torch.min(x2*Y32)
fig, (ax1) = plt.subplots(1,1) 
dls.train.decode((x_r,))[0].squeeze().show(ax=ax1)
ax1.set_title("ORIGINAL")
fig.set_figwidth(4)            
fig.set_figheight(4)
fig.tight_layout()

with open('CALTECH_101_random_hcam_cat_plt1.pkl', 'wb') as file:
    pickle.dump(fig, file)
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(x12*-1.5).squeeze().show(ax=ax1)  #MODE1
(x1*0.5).squeeze().show(ax=ax2)  #MODE1_res
ax1.set_title("MODE1")
ax2.set_title("MODE1 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
with open('CALTECH_101_random_hcam_cat_plt2.pkl', 'wb') as file:
    pickle.dump(fig, file)
    
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x22+1).squeeze().show(ax=ax1)  #MODE2
(x2*0.5).squeeze().show(ax=ax2)  #MODE2_res
ax1.set_title("MODE2")
ax2.set_title("MODE2 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()
with open('CALTECH_101_random_hcam_cat_plt3.pkl', 'wb') as file:
    pickle.dump(fig, file)
    
#
fig, (ax1, ax2) = plt.subplots(1,2) 
(-x32+1).squeeze().show(ax=ax1)  #MODE3
(x3*0.5).squeeze().show(ax=ax2)  #MODE3_res
ax1.set_title("MODE3")
ax2.set_title("MODE3 RES")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

with open('CALTECH_101_random_hcam_cat_plt4.pkl', 'wb') as file:
    pickle.dump(fig, file)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

with open('CALTECH_101_random_hcam_cat_plt1.pkl', 'rb') as file:
    hcam_cat_plt = pickle.load(file)
plt.show(hcam_cat_plt)

with open('CALTECH_101_random_hcam_cat_plt2.pkl', 'rb') as file:
    hcam_cat_plt = pickle.load(file)
plt.show(hcam_cat_plt)

with open('CALTECH_101_random_hcam_cat_plt3.pkl', 'rb') as file:
    hcam_cat_plt = pickle.load(file)
plt.show(hcam_cat_plt)

with open('CALTECH_101_random_hcam_cat_plt4.pkl', 'rb') as file:
    hcam_cat_plt = pickle.load(file)
plt.show(hcam_cat_plt)

fig, (ax1) = plt.subplots(1,1) 
(x12*-1.5).squeeze().show(ax=ax1)  #MODE1
ax1.set_title("MODE1")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

with open('CALTECH_101_random_hcam_cat_plt_mode1.pkl', 'wb') as file:
    pickle.dump(fig, file)

with open('CALTECH_101_random_hcam_cat_plt_mode1.pkl', 'rb') as file:
    hcam_cat_plt_mode1 = pickle.load(file)
plt.show(hcam_cat_plt_mode1)

fig, (ax1) = plt.subplots(1,1) 
(x12*-1.5 + -x22+1).squeeze().show(ax=ax1)  #MODE1+MODE2
ax1.set_title("MODE1+MODE2")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

with open('CALTECH_101_random_hcam_cat_plt_mode2.pkl', 'wb') as file:
    pickle.dump(fig, file)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

with open('CALTECH_101_random_hcam_cat_plt_mode2.pkl', 'rb') as file:
    hcam_cat_plt_mode2 = pickle.load(file)
plt.show(hcam_cat_plt_mode2)

fig, (ax1) = plt.subplots(1,1) 
(x12*-1.5 + -x22+1 + -x32+1).squeeze().show(ax=ax1)  #MODE1+MODE2+MODE3
ax1.set_title("MODE1+MODE2+MODE3")
fig.set_figwidth(8)            
fig.set_figheight(8)
fig.tight_layout()

with open('CALTECH_101_random_hcam_cat_plt_mode3.pkl', 'wb') as file:
    pickle.dump(fig, file)
Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

with open('CALTECH_101_random_hcam_cat_plt_mode3.pkl', 'rb') as file:
    hcam_cat_plt_mode3 = pickle.load(file)
plt.show(hcam_cat_plt_mode3)

Randombox Other Methods

x_r = x_r.to('cpu')

GradCAM

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

HiResCAM

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

ScoreCAM

scorecam_randombox = ScoreCAM(model=lrnr_r.model.to('cpu'), target_layers=lrnr_r.model[0][-1])
cam_scorecam_randombox = scorecam_randombox(input_tensor=x_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,  1.31it/s]
dls.train.decode((x_r,))[0].squeeze().show()
plt.imshow(cam_scorecam_randombox.squeeze(), alpha=0.7)

GradCAMPlusPlus

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

AblationCAM

ablationcam_randombox = AblationCAM(model=lrnr_r.model.to('cpu'), target_layers=lrnr_r.model[0][-1])
cam_ablationcam_randombox = ablationcam_randombox(input_tensor=x_r,targets=None)
100%|██████████| 16/16 [00:26<00:00,  1.64s/it]
100%|██████████| 16/16 [00:25<00:00,  1.58s/it]
100%|██████████| 16/16 [00:25<00:00,  1.57s/it]
dls.train.decode((x_r,))[0].squeeze().show()
plt.imshow(cam_ablationcam_randombox.squeeze(), alpha=0.7)

XGradCAM

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

EigenCAM

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

FullGrad

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

EigenGradCAM

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

LayerCAM

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

Figure_Randombox

fig, ((ax1,ax2,ax3,ax4),
      (ax5,ax6,ax7,ax8),
      (ax9,ax10,ax11,ax12)) = plt.subplots(3,4) 
plt.title('Randombox')
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax1)
ax1.imshow(cam_gradcam_randombox.squeeze(), alpha=0.7)
ax1.set_title("GradCAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax2)
ax2.imshow(cam_hirescam_randombox.squeeze(), alpha=0.7)
ax2.set_title("HiResCAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax3)
ax3.imshow(cam_scorecam_randombox.squeeze(), alpha=0.7)
ax3.set_title("ScoreCAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax4)
ax4.imshow(cam_gradcamplusplus_randombox.squeeze(), alpha=0.7)
ax4.set_title("GradCAMPlusPlus")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax5)
ax5.imshow(cam_ablationcam_randombox.squeeze(), alpha=0.7)
ax5.set_title("AblationCAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax6)
ax6.imshow(cam_xgradcam_randombox.squeeze(), alpha=0.7)
ax6.set_title("XGradCAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax7)
ax7.imshow(cam_eigencam_randombox.squeeze(), alpha=0.7)
ax7.set_title("EigenCAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax8)
ax8.imshow(cam_fullgrad_randombox.squeeze(), alpha=0.7)
ax8.set_title("FullGrad")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax9)
ax9.imshow(cam_eigengradcam_randombox.squeeze(), alpha=0.7)
ax9.set_title("EigenGradCAM")
#
dls.train.decode((x_r,))[0].squeeze().show(ax=ax10)
ax10.imshow(cam_layercam_randombox.squeeze(), alpha=0.7)
ax10.set_title("LayerCAM")
#
(dls.train.decode((x_r,))[0].squeeze()*0).show(ax=ax11)
(dls.train.decode((x_r,))[0].squeeze()*0).show(ax=ax12)
#
fig.set_figwidth(20)            
fig.set_figheight(20)
fig.tight_layout()

with open('CALTECH_101_random_fig_randombox_plt.pkl', 'wb') as file:
    pickle.dump(fig, file)

with open('CALTECH_101_random_fig_randombox_plt.pkl', 'rb') as file:
    fig_randombox_plt = pickle.load(file)
fig_randombox_plt.show()