class MLGUI:
def __init__(self, master):
self.master = master
master.title("My ML GUI")
self.network = MyNetwork()
self.label = Label(master, text="This is ML GUI")
self.label.pack()
self.load_button = Button(master, text="Load Data", command=self.load)
self.load_button.pack()
self.learn_button = Button(master, text="Teach Network", command=self.teach)
self.learn_button.pack()
self.test_button = Button(master, text="Test Network", command=self.test)
self.test_button.pack()
self.close_button = Button(master, text="Close", command=master.quit)
self.close_button.pack()
def load(self):
self.network.load_data()
print("Im loading")
def teach(self):
print("Im teaching")
self.network.learn()
def test(self):
print("Im testing")
predictions = self.network.predict()
show_image(predictions, self.network.test_labels, self.network.test_images)
def main(config):
"""The main function."""
# Initialize network
mynet = MyNetwork(config)
# Run propper mode
if config.run_mode == "train":
# Load data train and validation data
data = {}
data["train"] = load_data(config, "train")
data["valid"] = load_data(config, "valid")
# Run train
mynet.train(data)
elif config.run_mode == "test":
# Load validation and test data. Note that we also load validation to
# visualize more than what we did for training. For training we choose
# minimal reporting to not slow down.
data = {}
data["valid"] = load_data(config, "valid")
data["test"] = load_data(config, "test")
# Run train
mynet.test(data)
elif config.run_mode == "test_simple":
data = {}
data["valid"] = load_data(config, "valid")
data["test"] = load_data(config, "test")
mynet.test_simple(data)
elif config.run_mode == "comp":
# This mode is for running comparison experiments. While cleaning the
# code, I took many parts out to make the code cleaner, which may have
# caused some issues. We are releasing this part just to help
# researchers, but we will not provide any support for this
# part. Please use at your own risk.
data = {}
data["valid"] = load_data(config, "valid")
data["test"] = load_data(config, "test")
# Run train
mynet.comp(data)
def __init__(self, master):
self.master = master
master.title("My ML GUI")
self.network = MyNetwork()
self.label = Label(master, text="This is ML GUI")
self.label.pack()
self.load_button = Button(master, text="Load Data", command=self.load)
self.load_button.pack()
self.learn_button = Button(master, text="Teach Network", command=self.teach)
self.learn_button.pack()
self.test_button = Button(master, text="Test Network", command=self.test)
self.test_button.pack()
self.close_button = Button(master, text="Close", command=master.quit)
self.close_button.pack()
def model_generator():
l = [1000] + layers + [3]
network_layers = []
for i in range(1, len(l)):
network_layers.append(torch.nn.Linear(l[i - 1], l[i]))
if i != len(l) - 1:
network_layers.append(
get_activation_function(activation_function))
return MyNetwork(network_layers)
kp2 = kp2[sorted_idx, :]
scores = y.squeeze()[sorted_idx]
# filter out matches with negative scores
kp1 = kp1[scores > 0]
kp2 = kp2[scores > 0]
filtered_matches.append([kp1, kp2, im1_name, im2_name])
if save:
np.savez('feature_matches_filtered', filtered_matches=filtered_matches)
return np.array(filtered_matches)
IMG_PATHS_FILE = 'sfm_image_paths - Copy.txt'
# load unfiltered_matches:
# feature_matches = np.load('feature_matches_unfiltered.npz')['feature_matches']
# uncomment the following line if unfiltered matches do not exist
feature_matches = match_img_list(IMG_PATHS_FILE, save=True)
with open('./demo/config.pickle', 'rb') as handle:
config = pickle.load(handle)
mynet = MyNetwork(config)
mynet.load_model()
# generate filtered matched. Note matches are not cv.KeyPoint objects but 2D points
# example shape is filtered_matches[img_combinations[kps1, kps2[2D point]], img1_name, img_2 name]
filtered_matches = filter_matches(feature_matches, mynet, save=True)
from config import *
from network import MyNetwork, LOG, WARNING
def parse_network(net_str):
net_str = net_str.split('+')
net_backbone = net_str[:-1]
net_head = net_str[-1]
module = []
for b in net_backbone:
part = name2net[b](params, in_encoder=True)
if hasattr(part, 'output_channels'):
params.output_channels = part.output_channels
module.append(part)
module.append(name2net[net_head](params))
return module
network = 'resnet101+aspp_plus+deeplabv3_plus_decoder'
params = Params()
params.model = network
params.dataset_root = '/media/ubuntu/disk/cityscapes'
params.has_max_pool = False
params.output_stride = 16
net = MyNetwork(params, module=parse_network(network))
net.Train()
# coding: utf-8
import sys, os
sys.path.append(os.pardir) # 부모 디렉터리의 파일을 가져올 수 있도록 설정
import numpy as np
from network import MyNetwork
from matplotlib import pyplot as plt
# load data
x_train = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
t_train = np.array([[1, 0], [0, 1], [0, 1], [1, 0]])
x_test = np.array([[0, 0], [0, 1], [1, 0], [1, 1]])
t_test = np.array([[1, 0], [0, 1], [0, 1], [1, 0]])
network = MyNetwork(input_size=2, hidden_size=8, output_size=2)
iters_num = 5000
learning_rate = 0.1
train_size = x_train.shape[0]
batch_size = 4
iter_per_epoch = max(train_size / batch_size, 1)
train_loss_list = []
train_acc_list = []
test_acc_list = []
for i in range(iters_num):
# batch_mask = np.random.choice(train_size, batch_size)
# x_batch = x_train[batch_mask]
# t_batch = t_train[batch_mask]