changed path

3c2fa313 · Vedrana Andersen Dahl · b6c4df6d · 3c2fa313 · 3c2fa313 · 3c2fa313
Commit 3c2fa313 authored Feb 28, 2024 by Vedrana Andersen Dahl
--- a/Chapter05/circles_modeling_EMPTY.py
+++ b/Chapter05/circles_modeling_EMPTY.py
@@ -30,7 +30,7 @@ def segmentation_histogram(ax, D, S, edges=None):
        ax.plot(centers, np.histogram(D[S==k].ravel(), edges)[0])
-path = '../../data/week5/'
+path = '../data/week5/'
 D = skimage.io.imread(path + 'noisy_circles.png').astype(float)
 # Ground-truth segmentation.

--- a/Chapter05/circles_modelling_EMPTY.ipynb
+++ b/Chapter05/circles_modelling_EMPTY.ipynb
@@ -12,7 +12,7 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 79,
+   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
@@ -32,7 +32,7 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 80,
+   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
@@ -58,7 +58,7 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 81,
+   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
@@ -84,11 +84,11 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 82,
+   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
-    "path = '../../data/week5/'\n",
+    "path = '../data/week5/'\n",
    "D = skimage.io.imread(path + 'noisy_circles.png').astype(float)\n",
    "\n",
    "# Ground-truth segmentation.\n",
@@ -109,7 +109,7 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 83,
+   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
@@ -139,7 +139,7 @@
  },
  {
   "cell_type": "code",
-   "execution_count": 84,
+   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
@@ -196,6 +196,11 @@
   "cell_type": "markdown",
   "metadata": {},
   "source": []
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": []
  }
 ],
 "metadata": {

 %% Cell type:markdown id: tags:
 # Segmentation modelling
 The code provided in connection with the section 5.2 (MRF modelling for image segmentation) of the lecture notes.
 %% Cell type:code id: tags:
 ``` python
 import skimage.io
 import numpy as np
 import scipy.ndimage
 import matplotlib.pyplot as plt
 ```
 %% Cell type:markdown id: tags:
 You should write the code which computes segmentation energies (U1 and U2) as described in the lecture notes. The skeleton of the function is provided below.
 %% Cell type:code id: tags:
 ``` python
 def segmentation_energy(S, D, mu, beta):
    # TODO -- add your code here
     # likelihood energy
    U1 = 0
    # prior energy
    U2 = 0
    return U1, U2
 ```
 %% Cell type:markdown id: tags:
 A helping function for plotting the histograms of the data and the segmentation.
 %% Cell type:code id: tags:
 ``` python
 def segmentation_histogram(ax, D, S, edges=None):
    '''
    Plot histogram for grayscale data and each segmentation label.
    '''
    if edges is None:
        edges = np.linspace(D.min(), D.max(), 100)
    ax.hist(D.ravel(), bins=edges, color = 'k')
    centers = 0.5 * (edges[:-1] + edges[1:])
    for k in range(S.max() + 1):
        ax.plot(centers, np.histogram(D[S==k].ravel(), edges)[0])
 ```
 %% Cell type:markdown id: tags:
 Get hold of the data (noisy image) and ground truth segmentation.
 %% Cell type:code id: tags:
 ``` python
-path = '../../data/week5/'
+path = '../data/week5/'
 D = skimage.io.imread(path + 'noisy_circles.png').astype(float)
 # Ground-truth segmentation.
 GT = skimage.io.imread(path + 'noise_free_circles.png')
 (mu, S_gt) = np.unique(GT, return_inverse=True)
 S_gt = S_gt.reshape(D.shape)
 segmentations = [S_gt]  # list where I'll place different segmentations
 ```
 %% Cell type:markdown id: tags:
 Find some configurations (segmentations) using conventional segmentation methods.
 %% Cell type:code id: tags:
 ``` python
 # Simple thresholding
 S_t = np.zeros(D.shape, dtype=int) + (D > 100) + (D > 160) # thresholded
 segmentations += [S_t]
 # Gaussian filtering followed by thresholding
 D_s = scipy.ndimage.gaussian_filter(D, sigma=1, truncate=3, mode='nearest')
 S_g = np.zeros(D.shape, dtype=int) + (D_s > 100) + (D_s > 160)
 segmentations += [S_g]
 # Median filtering followed by thresholding
 D_m = scipy.ndimage.median_filter(D, size=(5, 5), mode='reflect')
 S_t = np.zeros(D.shape, dtype=int) + (D_m > 100) + (D_m > 160) # thresholded
 segmentations += [S_t]
 ```
 %% Cell type:markdown id: tags:
 Visualize the segmentations, associated histograms and error images. Once you implement the `segmentation_energy` you can look at whether there is a link between segmentation energies and the quality of the segmentation.
 %% Cell type:code id: tags:
 ``` python
 #%% visualization
 fig, ax = plt.subplots()
 ax.imshow(D, vmin=0, vmax=255, cmap=plt.cm.gray)
 plt.show()
 fig, ax = plt.subplots(3, len(segmentations), figsize=(10, 10))
 beta = 100
 for i, s in enumerate(segmentations):
    ax[0][i].imshow(s)
    U1, U2 = segmentation_energy(s, D, mu, beta)
    ax[0][i].set_title(f'likelihood: {int(U1)}\nprior: {U2}\nposterior: {int(U1)+U2}')
    segmentation_histogram(ax[1][i], D, s)
    ax[1][i].set_xlabel('Intensity')
    ax[1][i].set_ylabel('Count')
    err = S_gt - s
    ax[2][i].imshow(err, vmin=-2, vmax=2, cmap=plt.cm.bwr)
    ax[2][i].set_title(f'Pixel error: {(err != 0).sum()}')
 fig.tight_layout()
 plt.show()
 ```
 %% Output
 %% Cell type:markdown id: tags:
+%% Cell type:markdown id: tags:

--- a/Chapter05/gender_labeling.ipynb
+++ b/Chapter05/gender_labeling.ipynb