Add files via upload

Author: dwday

Example LBP transform on Cifar10 dataset.ipynb

@@ -1,8 +1,8 @@
 #!/usr/bin/env python3
 # -*- coding: utf-8 -*-
 """
-Example lbp application on Cifar10 
-
+Example LBP transform on Cifar10 dataset
+# Akgun, Devrim. "A PyTorch Operations Based Approach for Computing Local Binary Patterns." U. Porto Journal of Engineering 7.4 (2021): 61-69.
 https://journalengineering.fe.up.pt/index.php/upjeng/article/view/2183-6493_007-004_0005/567
 """
 from lib.lbplib import lbp_py,lbp_pt
@@ -12,6 +12,7 @@
 import torch
 import torchvision
 
+#Compute LBP using Python operations
 def py_extract_lbp_rgb(x_train):
     [N,Rows,Cols,Channels]=x_train.shape    
     x_train_lbp=np.zeros(shape=(N,Rows,Cols,Channels),dtype='uint8')    
@@ -22,7 +23,7 @@ def py_extract_lbp_rgb(x_train):
     return x_train_lbp
 
 
-
+#Compute LBP using PyTorch operations
 def torch_extract_lbp_rgb(x_train):
     [N,Rows,Cols,Channels]=x_train.shape
     x_train_lbp=np.zeros(shape=(N,Rows,Cols,Channels),dtype='uint8') 
@@ -31,31 +32,27 @@ def torch_extract_lbp_rgb(x_train):
     x_train_lbp[:,:,:,2]=lbp_pt(x_train[:,:,:,2]).numpy()
     return x_train_lbp
 
+#Use Cifar10 for the example
+trainset = torchvision.datasets.CIFAR10(root='./data', download=True)
 
-trainset = torchvision.datasets.CIFAR10(root='./data', train=True,
-                                        download=True)
 #use first 200 images for the test
 x_train=trainset.data[:200,:,:,:]
 
-# Process complete dataset 
-  
-# Extract lbp features for Cifar10 dataset using PyTorch
+# 1- Extract lbp features for the selected images using PyTorch ---------------
 start_time   = time.time() 
 
-#process all images in the Cifar10 --------------------------------------------
-
-
+# process all images using PyTorch
 x_train_pt =torch.from_numpy(x_train) 
 
 x_train_lbp_pt = torch_extract_lbp_rgb(x_train_pt) 
 
 elapsed_pt = time.time() - start_time
 print('PyTorch elapsed_time=',elapsed_pt)
 
-# Extract lbp features for Cifar10 dataset ------------------------------------
+# 2- Extract lbp features for the selected images using Python-----------------
 start_time   = time.time()
 
-#process all images in the Cifar10
+# process all images using Python
 x_train_lbp_py = py_extract_lbp_rgb(x_train) 
 
 elapsed_py = time.time() - start_time
@@ -67,7 +64,7 @@ def torch_extract_lbp_rgb(x_train):
 
 
 # Example images---------------------------------------------------------------
-#input images
+# Input images
 plt.figure(1)
 figs, axes = plt.subplots(4, 6)
 for i in range(4):
@@ -76,7 +73,7 @@ def torch_extract_lbp_rgb(x_train):
         axes[i, j].set_xticks([])
         axes[i, j].set_yticks([])
 
-# LBP images
+# LBP transformed images
 plt.figure(2)
 figs, axes = plt.subplots(4, 6)
 for i in range(4):

Example LBP transform using PyTorch.ipynb

@@ -5,10 +5,9 @@
 Test file for rgb and grascale outputs  
 https://journalengineering.fe.up.pt/index.php/upjeng/article/view/2183-6493_007-004_0005/567
 """
-from lib.lbplib import lbp_py,lbp_pt
+from lib.lbplib import lbp_pt
 from PIL import Image
 import matplotlib.pyplot as plt
-import time
 import numpy as np
 import torch
 

example_cifar10_processing.py .py

@@ -44,5 +44,5 @@
 
 print('python:')
 print(ypy)
-print('TensorFlow:')
+print('PyTorch:')
 print(ypt[0,:,:])    

example_image_rgb.py

@@ -8,7 +8,9 @@
 #import time
 #os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
 print("torch.cuda.is_available():",torch.cuda.is_available())
-#------------------------------------------------------------------------------
+
+
+#- Python implementation of LBP transform -------------------------------------
 def lbp_python(Im): 
 
     sat=len(Im)
@@ -29,10 +31,10 @@ def lbp_python(Im):
             ( I[i-1,j-1]>= I[i,j] )*128;  
 
     return L
-#------------------------------------------------------------------------------
 
 
-def tc_lbp(x):
+#- PyTorch implementation of LBP transform ------------------------------------
+def lbp_torch(x):
     #pad image for 3x3 mask size
     x = F.pad(input=x, pad = [1, 1, 1, 1], mode='constant')
     b=x.shape
@@ -91,7 +93,7 @@ def tc_lbp(x):
 print(imgs)
 
 # Compute using pytorch
-y1=tc_lbp(torch.from_numpy(imgs.reshape(1,7,7)))
+y1=lbp_torch(torch.from_numpy(imgs.reshape(1,7,7)))
 # Compute using python
 y2=lbp_python(imgs)
 

example_test.py

@@ -0,0 +1,6 @@
+matplotlib==3.6.0
+numpy==1.23.3
+Pillow==9.2.0
+Pillow==10.0.1
+torch==1.12.0+cu113
+torchvision==0.13.0+cu113