adriaciurana

import numpy as np
import matplotlib.pyplot as plt
from matplotlib.animation import FuncAnimation
from mpl_toolkits.mplot3d import Axes3D

# numero iteraciones
ITERS = 10000

# learning rate
LR = 0.0000000001

adriaciurana

A = roty(-30); B = 2; C = rotx(45);
[A2, B2, C2] = svd(A*B*C);

if(sum(sum(abs(A2)-abs(A))) < 1e-10 && ...
        sum(sum(abs(B2)-abs(B))) < 1e-10 && ...
        sum(sum(abs(C2)-abs(C))) < 1e-10)
    disp('La descomposicion da como resultado A, B, C iguales en magnitud');
end

adriaciurana

adriaciurana

import argparse, keras, os
from keras.models import model_from_json
parser = argparse.ArgumentParser()
parser.add_argument("image", help="Image or folder of images to predict",
                    type=str)
args = parser.parse_args()

# Cargamos el modelo
json_file = open('model.json', 'r')
loaded_model_json = json_file.read()

adriaciurana

import argparse, keras, os, cv2
from keras.models import model_from_json
parser = argparse.ArgumentParser()
parser.add_argument("image", help="Image or folder of images to predict",
                    type=str)
args = parser.parse_args()

adriaciurana

# Cargamos el modelo
json_file = open('model.json', 'r')
loaded_model_json = json_file.read()
json_file.close()
model = model_from_json(loaded_model_json)
model.load_weights("model.h5")

adriaciurana

# Cargamos la imagen/imagenes
if os.path.isdir(args.image):
	path_images = []
	for ext in ['jpg', 'jpeg', 'tiff', 'png', 'gif']:
		path_images += glob.glob(os.path.join(args.image, "*." + ext))
else:
	path_images = [args.image]

images = []
for path_im in path_images:

adriaciurana

Epoch 50/50
...
2000/2000 [==============================] - 158s 79ms/step - loss: 0.1001 - binary_accuracy: 0.9613 - val_loss: 0.6304 - val_binary_accuracy: 0.8426

adriaciurana

# Cargamos la clase para generar modelos sequenciales
from keras.models import Sequential 
# Cargamos las siguientes capas
from keras.layers import Conv2D, MaxPooling2D, Dense, Flatten

model = Sequential()
# Bloque 1:
# - Conv1a: neuronas=128, ventana=(3,3), activacion=ReLU.
# - Conv1b: neuronas=256, ventana=(3,3), activacion=ReLU.
# - Max-pooling: ventana=(3,3), stride=2.

adriaciurana

# Usamos el optimizador Adam
# Usamos la loss function binary_crossentropy. Esta es util cuando la red devuelve una probabilidad, es este caso ser perro o ser gato.
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['binary_accuracy'])
model.summary()