def predict(path, debug=False):
global net, image_size
# img = load_image(path)
X = np.asarray([path])
img = np.asarray([load_image(path)], dtype="float32")
cats = load_classes()
probs = net.predict_proba(X)
if debug == True:
for layer in net.layers:
name = layer[0].__name__
filename = os.path.splitext(path)[0]
try:
os.mkdir("debug/" + filename)
except:
pass
if "2D" in name:
real_name = layer[1]["name"]
try:
plt = plot_conv_activity(net.layers_[real_name], img)
plt.savefig("debug/" + filename + "/" + real_name +
"_activity.png")
plt.close()
#plt = plot_conv_weights(net.layers_[real_name]) DISABLED: Eats too much memory, pretty useless anyhow
#plt.savefig("debug/"+filename+"/"+real_name+"_weights.png")
except Exception as e:
print(e)
for i, x in enumerate(probs[0]):
print(cats[i].name, x)
result = [(cats[np.argmax(x)].id, cats[np.argmax(x)].name, x[np.argmax(x)])
for x in probs]
return result[0]
def load_data(converter, test=False, return_files=False):
images = []
labels = []
imgs = []
if test:
print("Loading test data...")
else:
print("Loading data...")
cats = data.load_classes()
for cat in cats:
print("Loading category {0}...".format(cat.name))
data.delete_broken_images(cat.name)
path = cat.name
if test:
path = "g_" + path
valid_images = [".jpg", ".gif", ".png", ".tga", ".jpeg"]
for f in os.listdir(path):
ext = os.path.splitext(f)[1]
if ext.lower() not in valid_images:
continue
img = load_image(os.path.join(path, f), converter)
imgs.append(os.path.join(path, f))
images.append(img)
labels.append(cat.name)
print("Loaded data.")
if return_files:
return images, labels, imgs
return images, labels
def plot_occ(path, real):
global net
img = load_image(path)
cats = load_classes()
X = np.asarray([img], dtype="float32")
plt = plot_occlusion(net, X, [real])
return plt
def convert_num_topics(info, num_topics):
if isinstance(num_topics, str):
gt = len(data.load_classes(info))
if num_topics == "gt":
num_topics = gt
elif num_topics.endswith("gt"):
num_topics = int(float(num_topics[:-2]) * gt)
else:
raise config.ConfigException(
'Unknown "num_topics" format: {}'.format(num_topics))
return num_topics
def score_exceptions():
global net
X_test, y_test = load_g_data()
cats = load_classes()
res = net.predict(X_test)
count = 0
for r, x, y in zip(res, X_test, y_test):
if r != y:
print(x)
copyfile(
x, "exceptions/predicted_" + cats[r].name + "_actual_" +
cats[y].name + "_" + str(count) + os.path.splitext(x)[1])
count += 1
from nolearn.lasagne.handlers import SaveWeights
from nolearn.lasagne.visualize import draw_to_file, plot_occlusion, plot_conv_activity, plot_conv_weights, draw_to_notebook
from skimage.io import imread
from skimage.color import gray2rgb, rgba2rgb
from skimage.transform import resize
from nolearn_utils.iterators import (BufferedBatchIteratorMixin,
ShuffleBatchIteratorMixin,
AffineTransformBatchIteratorMixin,
ReadImageBatchIteratorMixin,
RandomFlipBatchIteratorMixin,
make_iterator)
from nolearn_utils.hooks import (SaveTrainingHistory, PlotTrainingHistory)
cats = load_classes()
image_size = 224
batch_size = 128
n_classes = len(cats) #DONE:0 Determine number of classes
print(n_classes)
train_iterator_mixins = [
ShuffleBatchIteratorMixin,
ReadImageBatchIteratorMixin,
RandomFlipBatchIteratorMixin,
AffineTransformBatchIteratorMixin,
BufferedBatchIteratorMixin,
]
TrainIterator = make_iterator('TrainIterator', train_iterator_mixins)
from __future__ import division
import os
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
import numpy as np
import cv2
from models.darknet import Darknet
from utils import *
import os
from data import load_classes, prep_image
test_dir = 'data/test_imgs'
class_names = load_classes("data/coco.names")
image_names = [os.path.join(test_dir, n_) for n_ in os.listdir(test_dir)]
images = list(map(cv2.imread, image_names))
input_images = [prep_image(im_, 416) for im_ in images]
dnet = Darknet("cfg/yolov3.cfg")
weight_file = 'checkpoints/yolov3.weights'
dnet.load_net_binary(weight_file)
dnet.eval()
#
l_0 = nn.Sequential(list(dnet.module_list)[0][:2]) # first conv layer
h = l_0(input_images[0])
h0 = torch.load(os.path.expanduser("~/tmp/c0a.pt"))
print((h-h0).abs().max())
print('')