[CAM]HCAM Tutorial

Author

SEOYEON CHOI

Published

November 29, 2023

Import

import HCAM
from torchvision.models import *
from fastai.vision.all import *

Image Loader

dls = HCAM.Img_loader(fil_path = 'random_pet_one',resize = 512)

CAM

lrnr=cnn_learner(dls,resnet34,metrics=error_rate)
lrnr.fine_tune(1)

epoch	train_loss	valid_loss	error_rate	time
0	0.124668	0.003786	0.000677	00:33

epoch	train_loss	valid_loss	error_rate	time
0	0.002606	0.000005	0.000000	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(1)

epoch	train_loss	valid_loss	accuracy	time
0	0.014341	0.000901	1.000000	00:43

epoch	train_loss	valid_loss	accuracy	time
0	0.000014	0.000002	1.000000	00:43

x_cat, = first(dls.test_dl([PILImage.create(get_image_files(Path('random_pet_one'))[2])]))
x_cat = x_cat.to('cpu')

HCAM

Mode 1

one = HCAM.HCAM(lrnr = lrnr2)

one.learner_thresh(Thresh=1600,input_img=x_cat)

/home/csy/Dropbox/blog/posts/CAM/HCAM/learners.py:66: 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)))
/home/csy/Dropbox/blog/posts/CAM/HCAM/learners.py:67: 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).
  self.ybar_threshed = np.where(power_threshed>Thresh,torch.tensor(camimg[0].detach().reshape(-1)),0)

one.learner_step(Rate=-0.05)

one.prob(input_img=x_cat)

one.mode_decomp(input_img=x_cat)

# one(input_img=x_cat)

HCAM.plot(dls,input_img=x_cat,
         input_img1=one(input_img=x_cat)['x'],input_img1_res=one(input_img=x_cat)['x_res'],
         one=0.3, one_res=0.2)

Clipping input data to the valid range for imshow with RGB data ([0..1] for floats or [0..255] for integers).

Mode 2

one.learner_thresh(Thresh=1600,input_img=one(input_img=x_cat)['x_res'])

one.learner_step(Rate=-0.05)

one.prob(input_img=one(input_img=x_cat)['x_res'])

one.mode_decomp(input_img=one(input_img=x_cat)['x_res'])

# one(input_img=one(input_img=x_cat)['x_res'])

HCAM.plot(dls,input_img=x_cat,
         input_img1=one(input_img=x_cat)['x'],input_img1_res=one(input_img=x_cat)['x_res'],
         input_img2=one(input_img=one(input_img=x_cat)['x'])['x'],input_img2_res=one(input_img=one(input_img=x_cat)['x_res'])['x_res'],
         one=0.35, one_res=0.2, two=0.5, two_res=0.2)

Other Methods

gradcam = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='gradcam',input_img=x_cat)

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(gradcam.squeeze(), alpha=0.7)

hirescam = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='hirescam',input_img=x_cat)

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(hirescam.squeeze(), alpha=0.7)

gradcamplusplus = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='gradcamplusplus',input_img=x_cat)

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(gradcamplusplus.squeeze(), alpha=0.7)

ablationcam = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='ablationcam',input_img=x_cat)

100%|██████████| 16/16 [00:27<00:00,  1.69s/it]
100%|██████████| 16/16 [00:27<00:00,  1.70s/it]
100%|██████████| 16/16 [00:26<00:00,  1.67s/it]

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(ablationcam.squeeze(), alpha=0.7)

xgradcam = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='xgradcam',input_img=x_cat)

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(xgradcam.squeeze(), alpha=0.7)

eigencam = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='eigencam',input_img=x_cat)

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(eigencam.squeeze(), alpha=0.7)

fullgrad = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='fullgrad',input_img=x_cat)

Warning: target_layers is ignored in FullGrad. All bias layers will be used instead

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(fullgrad.squeeze(), alpha=0.7)

eigengradcam = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='eigengradcam',input_img=x_cat)

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(eigengradcam.squeeze(), alpha=0.7)

layercam = HCAM.Other_method(lrnr=lrnr2, status='cpu', cam_method='layercam',input_img=x_cat)

dls.train.decode((x_cat,))[0].squeeze().show()
plt.imshow(layercam.squeeze(), alpha=0.7)