1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
|
{
"nbformat_minor": 0,
"nbformat": 4,
"cells": [
{
"execution_count": null,
"cell_type": "code",
"source": [
"%matplotlib inline"
],
"outputs": [],
"metadata": {
"collapsed": false
}
},
{
"source": [
"\n========================================================\nOT for domain adaptation with image color adaptation [6]\n========================================================\n\n[6] Ferradans, S., Papadakis, N., Peyre, G., & Aujol, J. F. (2014). Regularized discrete optimal transport. SIAM Journal on Imaging Sciences, 7(3), 1853-1882.\n\n"
],
"cell_type": "markdown",
"metadata": {}
},
{
"execution_count": null,
"cell_type": "code",
"source": [
"import numpy as np\nimport scipy.ndimage as spi\nimport matplotlib.pylab as pl\nimport ot\n\n\n#%% Loading images\n\nI1=spi.imread('../data/ocean_day.jpg').astype(np.float64)/256\nI2=spi.imread('../data/ocean_sunset.jpg').astype(np.float64)/256\n\n#%% Plot images\n\npl.figure(1)\n\npl.subplot(1,2,1)\npl.imshow(I1)\npl.title('Image 1')\n\npl.subplot(1,2,2)\npl.imshow(I2)\npl.title('Image 2')\n\npl.show()\n\n#%% Image conversion and dataset generation\n\ndef im2mat(I):\n \"\"\"Converts and image to matrix (one pixel per line)\"\"\"\n return I.reshape((I.shape[0]*I.shape[1],I.shape[2]))\n\ndef mat2im(X,shape):\n \"\"\"Converts back a matrix to an image\"\"\"\n return X.reshape(shape)\n\nX1=im2mat(I1)\nX2=im2mat(I2)\n\n# training samples\nnb=1000\nidx1=np.random.randint(X1.shape[0],size=(nb,))\nidx2=np.random.randint(X2.shape[0],size=(nb,))\n\nxs=X1[idx1,:]\nxt=X2[idx2,:]\n\n#%% Plot image distributions\n\n\npl.figure(2,(10,5))\n\npl.subplot(1,2,1)\npl.scatter(xs[:,0],xs[:,2],c=xs)\npl.axis([0,1,0,1])\npl.xlabel('Red')\npl.ylabel('Blue')\npl.title('Image 1')\n\npl.subplot(1,2,2)\n#pl.imshow(I2)\npl.scatter(xt[:,0],xt[:,2],c=xt)\npl.axis([0,1,0,1])\npl.xlabel('Red')\npl.ylabel('Blue')\npl.title('Image 2')\n\npl.show()\n\n\n\n#%% domain adaptation between images\n\n# LP problem\nda_emd=ot.da.OTDA() # init class\nda_emd.fit(xs,xt) # fit distributions\n\n\n# sinkhorn regularization\nlambd=1e-1\nda_entrop=ot.da.OTDA_sinkhorn()\nda_entrop.fit(xs,xt,reg=lambd)\n\n\n\n#%% prediction between images (using out of sample prediction as in [6])\n\nX1t=da_emd.predict(X1)\nX2t=da_emd.predict(X2,-1)\n\n\nX1te=da_entrop.predict(X1)\nX2te=da_entrop.predict(X2,-1)\n\n\ndef minmax(I):\n return np.minimum(np.maximum(I,0),1)\n\nI1t=minmax(mat2im(X1t,I1.shape))\nI2t=minmax(mat2im(X2t,I2.shape))\n\nI1te=minmax(mat2im(X1te,I1.shape))\nI2te=minmax(mat2im(X2te,I2.shape))\n\n#%% plot all images\n\npl.figure(2,(10,8))\n\npl.subplot(2,3,1)\n\npl.imshow(I1)\npl.title('Image 1')\n\npl.subplot(2,3,2)\npl.imshow(I1t)\npl.title('Image 1 Adapt')\n\n\npl.subplot(2,3,3)\npl.imshow(I1te)\npl.title('Image 1 Adapt (reg)')\n\npl.subplot(2,3,4)\n\npl.imshow(I2)\npl.title('Image 2')\n\npl.subplot(2,3,5)\npl.imshow(I2t)\npl.title('Image 2 Adapt')\n\n\npl.subplot(2,3,6)\npl.imshow(I2te)\npl.title('Image 2 Adapt (reg)')\n\npl.show()"
],
"outputs": [],
"metadata": {
"collapsed": false
}
}
],
"metadata": {
"kernelspec": {
"display_name": "Python 2",
"name": "python2",
"language": "python"
},
"language_info": {
"mimetype": "text/x-python",
"nbconvert_exporter": "python",
"name": "python",
"file_extension": ".py",
"version": "2.7.12",
"pygments_lexer": "ipython2",
"codemirror_mode": {
"version": 2,
"name": "ipython"
}
}
}
}
|
