Myosotis-Researches 0.1.40__tar.gz → 0.1.42__tar.gz

{myosotis_researches-0.1.40 → myosotis_researches-0.1.42/Myosotis_Researches.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: Myosotis-Researches
-Version: 0.1.40
+Version: 0.1.42
 Summary: A repository for storing my progress of researches.
 Home-page: https://github.com/Zeyu-Xie/Myosotis-Researches
 Author: Zeyu Xie

{myosotis_researches-0.1.40/Myosotis_Researches.egg-info → myosotis_researches-0.1.42}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: Myosotis-Researches
-Version: 0.1.40
+Version: 0.1.42
 Summary: A repository for storing my progress of researches.
 Home-page: https://github.com/Zeyu-Xie/Myosotis-Researches
 Author: Zeyu Xie

{myosotis_researches-0.1.40 → myosotis_researches-0.1.42}/myosotis_researches/CcGAN/utils/similarity.py

@@ -2,6 +2,7 @@ import cv2
 import math
 import numpy as np
 from PIL import Image
+from scipy.special import rel_entr
 from skimage.metrics import structural_similarity as ssim
 
 
@@ -92,12 +93,14 @@ def cross_correlation(images):
 def _rgb_histogram(image, bins=32):
     """
     image: numpy array of shape (H, W, 3), RGB format
-    returns: concatenated histogram for R, G, B channels
+    returns: concatenated and normalized histogram for R, G, B channels
     """
-    hist_r, _ = np.histogram(image[:, :, 0], bins=bins, range=(0, 255))
-    hist_g, _ = np.histogram(image[:, :, 1], bins=bins, range=(0, 255))
-    hist_b, _ = np.histogram(image[:, :, 2], bins=bins, range=(0, 255))
-    hist = np.concatenate([hist_r, hist_g, hist_b])
+    image = np.asarray(image)
+    hist_r, _ = np.histogram(image[:, :, 0], bins=bins, range=(0, 256))
+    hist_g, _ = np.histogram(image[:, :, 1], bins=bins, range=(0, 256))
+    hist_b, _ = np.histogram(image[:, :, 2], bins=bins, range=(0, 256))
+    hist = np.concatenate([hist_r, hist_g, hist_b]).astype(np.float64)
+    hist /= np.sum(hist)  # Normalize to make it a probability distribution
     return hist
 
 
@@ -118,6 +121,33 @@ def histogram_intersection(images):
     return histogram_intersection_sum / (n * (n - 1) / 2)
 
 
+# KL divergence
+def _kl_divergence(image_1, image_2, bins=256):
+    kl_values = []
+
+    for c in range(3):
+        arr1 = np.array(image_1)[:, :, c].flatten()
+        arr2 = np.array(image_2)[:, :, c].flatten()
+        hist1, _ = np.histogram(arr1, bins=bins, range=(0, 256), density=True)
+        hist2, _ = np.histogram(arr2, bins=bins, range=(0, 256), density=True)
+        hist1 += 1e-10
+        hist2 += 1e-10
+        kl_1_2 = np.sum(rel_entr(hist1, hist2))
+        kl_2_1 = np.sum(rel_entr(hist2, hist1))
+        sym_kl = 0.5 * (kl_1_2 + kl_2_1)
+        kl_values.append(sym_kl)
+    return float(np.mean(kl_values))
+
+
+def kl_divergence(images):
+    kl_divergence_sum = 0
+    n = len(images)
+    for i in range(n):
+        for j in range(i + 1, n):
+            kl_divergence_sum += _kl_divergence(images[i], images[j])
+    return kl_divergence_sum / (n * (n - 1) / 2)
+
+
 # Similarity
 _method_dict = {
     "MSE": MSE,
@@ -125,6 +155,7 @@ _method_dict = {
     "PSNR": PSNR,
     "cross-correlation": cross_correlation,
     "histogram-intersection": histogram_intersection,
+    "kl-divergence": kl_divergence,
 }
 
 

{myosotis_researches-0.1.40 → myosotis_researches-0.1.42}/setup.py

@@ -2,7 +2,7 @@ from setuptools import setup, find_packages
 
 setup(
     name="Myosotis-Researches",
-    version="0.1.40",
+    version="0.1.42",
     description="A repository for storing my progress of researches.",
     long_description=open("README.md").read(),
     long_description_content_type="text/markdown",