blog – untitled-copy1

from fastai.vision.all import *

path = untar_data(URLs.IMAGENETTE_320)

img = array(Image.open((path/'train').ls()[0].ls()[0]))

show_image(img),img.shape

(<AxesSubplot:>, (320, 463, 3))

Schedule: Question, This notebook, How to read a research paper, Presentation/Less formal

np.random.choice(np.arange(14),8,replace=False)

array([2, 5, 1, 0, 9, 8, 6, 7])

img.shape

(320, 463, 3)

img.shape[0]//2,img.shape[1]//2

(160, 231)

ex=img[:img.shape[0]//2,:img.shape[1]//2]
show_image(ex)

<AxesSubplot:>

img.shape[0]//4,-img.shape[0]//4

(80, -80)

ex=img[img.shape[0]//4:-img.shape[0]//4,
       img.shape[1]//4:-img.shape[1]//4]
show_image(ex)

<AxesSubplot:>

ex.shape,img.shape

((160, 232, 3), (320, 463, 3))

show_image(img)

<AxesSubplot:>

list(range(10,0,-1))

[10, 9, 8, 7, 6, 5, 4, 3, 2, 1]

show_image(img),img.shape

(<AxesSubplot:>, (320, 463, 3))

show_image(img[:-1:2,:-1:2]),img[:-1:2,:-1:2].shape

(<AxesSubplot:>, (160, 231, 3))

show_image(img[1::2,1::2]),img[1::2,1::2].shape

(<AxesSubplot:>, (160, 231, 3))

img[:-1:2,:-1:2]//2+img[1::2,1::2]//2

img[::2,::2]//2+img[1::2,1::2]//2

ValueError: operands could not be broadcast together with shapes (160,232,3) (160,231,3)

img[1::2,1::2].shape

(160, 231, 3)

ex= img[:-1:2,:-1:2]//2+img[1::2,1::2]//2
show_image(ex),ex.shape

(<AxesSubplot:>, (160, 231, 3))

array([1/3,1/3,1/3])

array([0.33333333, 0.33333333, 0.33333333])

(array([1/3,1/3,1/3])@img[...,None]).shape

(320, 463, 1)

ex1.max(),ex2.max()

(765, 255.0)

np.

ex1= (array([1/2,1/2,1/2])@img[...,None]).clip(0,255)
ex2= (array([1/3,1/3,1/3])@img[...,None]).clip(0,255)
show_image(ex1,cmap='gray'),show_image(ex2,cmap='gray')

(<AxesSubplot:>, <AxesSubplot:>)

norm_tfm=Normalize.from_stats(*imagenet_stats,cuda=False)
def show_norm(img): show_images((norm_tfm.decode(img).clamp(0,1)),nrows=3)

norm_img = norm_tfm(TensorImage(img.transpose(2,0,1)).float()[None]/255)

noise= torch.randn_like(norm_img)

As = torch.linspace(0,1,12)[...,None,None,None]; As.squeeze()

tensor([0.0000, 0.0909, 0.1818, 0.2727, 0.3636, 0.4545, 0.5455, 0.6364, 0.7273,
        0.8182, 0.9091, 1.0000])

(As)**.5*norm_img

(1-As**.5).squeeze()

tensor([1.0000, 0.6985, 0.5736, 0.4778, 0.3970, 0.3258, 0.2615, 0.2023, 0.1472,
        0.0955, 0.0465, 0.0000])

((1-As)**.5).squeeze()

tensor([1.0000, 0.9535, 0.9045, 0.8528, 0.7977, 0.7385, 0.6742, 0.6030, 0.5222,
        0.4264, 0.3015, 0.0000])

show_norm((As)**.5*norm_img+(1-As)**.5*noise)

show_norm((As)**.5*norm_img+(1-As**.5)*noise)

As.squeeze(),As.shape

(tensor([0.0000, 0.0909, 0.1818, 0.2727, 0.3636, 0.4545, 0.5455, 0.6364, 0.7273,
         0.8182, 0.9091, 1.0000]),
 torch.Size([12, 1, 1, 1]))

norm_img.shape

torch.Size([1, 3, 320, 463])

show_norm((As)**.5*norm_img)

1-(As)**.5

tensor([[[[1.0000]]],


        [[[0.6985]]],


        [[[0.5736]]],


        [[[0.4778]]],


        [[[0.3970]]],


        [[[0.3258]]],


        [[[0.2615]]],


        [[[0.2023]]],


        [[[0.1472]]],


        [[[0.0955]]],


        [[[0.0465]]],


        [[[0.0000]]]])

show_norm((1-As)**.5*noise)