def decafImages(src_path, socketid, output_path, result_path):
# Establishing connection to send results and write messages
print "The worker is working properly"
from skimage import io
from cloudcv17 import config
import redis
import json
import caffe
import numpy as np
import os
import glob
import time
import scipy.io as sio
import traceback
rs = redis.StrictRedis(host=config.REDIS_HOST, port=6379)
try:
# Needed to fix error https://github.com/BVLC/caffe/issues/438
io.use_plugin('matplotlib')
# Caffe Initialisations
CAFFE_DIR = os.path.normpath(os.path.join(os.path.dirname(caffe.__file__), "..", ".."))
MODEL_FILE = os.path.join(CAFFE_DIR, 'models/bvlc_reference_caffenet/deploy.prototxt')
PRETRAINED = os.path.join(CAFFE_DIR, 'models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel')
# Set CPU mode
caffe.set_mode_cpu()
# Make classifier.
classifier = caffe.Classifier(MODEL_FILE, PRETRAINED)
# Find decaf features and send Results
if os.path.isdir(src_path):
for input_file in glob.glob(os.path.join(src_path, '*')):
if os.path.isfile(input_file):
rs.publish('chat', json.dumps({'message': 'Processing ' +
os.path.basename(input_file), 'socketid': str(socketid)}))
# Loading Image
input_image = caffe.io.load_image(input_file)
# Finding decaf features
start = time.time()
classifier.predict([input_image])
blobs = classifier.blobs.items()
features = blobs[-3][1].data[:, :, 0, 0]
features_center = blobs[-3][1].data[4, :, 0, 0]
features_center = np.resize(features_center, (1, 4096))
timeMsg = "Completed in %.2f s." % (time.time() - start)
rs.publish('chat', json.dumps({'message': timeMsg, 'socketid': str(socketid)}))
# Saving decaf features
matfile = {}
matfile['decaf'] = features
matfile['decaf_center'] = features_center
out_file = os.path.join(output_path, os.path.basename(input_file) + '.mat')
publish_file = os.path.join(result_path, os.path.basename(input_file) + '.mat')
sio.savemat(out_file, matfile)
rs.publish('chat', json.dumps({'web_result': publish_file, 'socketid': str(socketid)}))
else:
input_file = src_path
rs.publish('chat', json.dumps({'message': 'Processing ' +
os.path.basename(input_file), 'socketid': str(socketid)}))
# Loading Image
input_image = caffe.io.load_image(input_file)
# Finding decaf features
start = time.time()
classifier.predict([input_image])
blobs = classifier.blobs.items()
features = blobs[-3][1].data[:, :, 0, 0]
features_center = blobs[-3][1].data[4, :, 0, 0]
features_center = np.resize(features_center, (1, 4096))
timeMsg = "Completed in %.2f s." % (time.time() - start)
rs.publish('chat', json.dumps({'message': timeMsg, 'socketid': str(socketid)}))
# Saving decaf features
matfile = {}
matfile['decaf'] = features
matfile['decaf_center'] = features_center
out_file = os.path.join(result_path, os.path.basename(input_file) + '.mat')
out_file = os.path.join(output_path, os.path.basename(input_file) + '.mat')
publish_file = os.path.join(result_path, os.path.basename(input_file) + '.mat')
sio.savemat(out_file, matfile)
rs.publish('chat', json.dumps({'web_result': publish_file, 'socketid': str(socketid)}))
rs.publish('chat', json.dumps({'message': 'Thank you for using CloudCV', 'socketid': str(socketid)}))
except:
# In case of an error, send the whole error with traceback
rs.publish('chat', json.dumps({'message': str(traceback.format_exc()), 'socketid': str(socketid)}))
import caffe
from caffe import layers as L, params as P
path = '/home/User/Documents/folder/'
#%%
#load net
model = path + 'real_deploy.prototxt'
weights = path + 'snap_iter_1000.caffemodel'
caffe.set_mode_cpu()
#caffe.set_device(0) #per gpu
net = caffe.Classifier(model, weights, raw_scale=1, image_dims=(128, 128))
#%%
#load test dataset from hdf5 file
with h5py.File(path + 'test.h5', 'r') as hf:
dataset = hf.get('data')
labelset = hf.get('label')
data = np.array(dataset)
label = np.array(labelset)
#compute predictions for every image in the test dataset
predictions = np.zeros(label.shape)
for i in range(data.shape[0]):
input_image = np.expand_dims(data[i, 0], axis=3)
import skimage
import requests
MODLE_FILE = "/home/yonatan/trendi/yonatan/Alexnet_deploy.prototxt"
PRETRAINED = "/home/yonatan/alexnet_imdb_first_try/caffe_alexnet_train_faces_iter_10000.caffemodel"
caffe.set_mode_gpu()
image_dims = [115, 115]
mean, input_scale = np.array([120, 120, 120]), None
channel_swap = [2, 1, 0]
raw_scale = 255.0
# Make classifier.
classifier = caffe.Classifier(MODLE_FILE,
PRETRAINED,
image_dims=image_dims,
mean=mean,
input_scale=input_scale,
raw_scale=raw_scale,
channel_swap=channel_swap)
print "Done initializing!"
def cv2_image_to_caffe(image):
return skimage.img_as_float(cv2.cvtColor(image, cv2.COLOR_BGR2RGB)).astype(
np.float32)
def url_to_image(url):
# download the image, convert it to a NumPy array, and then read
# it into OpenCV format
def classifynum(correctphone, bmpnum):
MODEL_FILE = '/home/songqing/dl/caffe/examples/mnist/lenet.prototxt'
PRETRAINED = '/home/songqing/dl/caffe/examples/mnist/zsq11_lenet_iter_10000.caffemodel'
piccount = 0
iternum = 0
finallist = []
poslist = []
ratelist = []
outfile = open('smallpic.txt', 'a+')
outfile.write('phone ' + str(bmpnum) + '\n')
outfile.write(str(correctphone) + '\n')
for iternum in range(1, 140):
#print iternum
counttemp = piccount
piccount = piccount + 1
filename = '0'
if counttemp < 10:
filename = filename + '00' + str(counttemp)
elif counttemp < 100:
filename = filename + '0' + str(counttemp)
else:
filename = filename + str(counttemp)
IMAGE_FILE = '/home/songqing/dl/dl/phone/phonenum_recognition/BMP7/testpic/pic' + filename + '.bmp'
if (os.path.isfile(IMAGE_FILE) == False):
continue
caffe.set_phase_test()
caffe.set_mode_cpu()
net = caffe.Classifier(MODEL_FILE, PRETRAINED)
# mean=np.load(caffe_root + 'python/caffe/imagenet/ilsvrc_2012_mean.npy'),
# channel_swap=(2,1,0),
# raw_scale=28,
# image_dims=(28, 28))
input_image = caffe.io.load_image(IMAGE_FILE, color=False)
#plt.imshow(input_image)
prediction = net.predict([input_image], oversample=False)
nummax = prediction[0].argmax()
#print 'prediction shape:', prediction[0].shape
#plt.plot(prediction[0])
#print 'predicted class:', prediction[0].argmax()
#print 'predicted class:', prediction[0][nummax]
#print 'predicted class:', prediction[0]
# if rate is small, put 0 to it's value
if (prediction[0][nummax] < 0.98 and nummax != 1):
if (nummax == 9):
if (prediction[0][nummax] < 0.9):
prediction[0][nummax] = 0
else:
prediction[0][nummax] = 0
if (prediction[0][nummax] < 0.995 and nummax == 1):
prediction[0][nummax] = 0
if (prediction[0][nummax] == 0):
outfile.write(
str(nummax) + ' ' + str(prediction[0][nummax]) + '\n')
continue
#in a range , get the max value
if ((len(finallist) == 0)):
if (nummax == 1): # the first num is 1
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
elif (iternum - poslist[-1] <= 2):
if (prediction[0][nummax] > ratelist[-1]):
if (len(finallist) > 1):
if (iternum - poslist[-2] < 14):
if (nummax == 1):
if (prediction[0][nummax] > 0.999):
finallist[-1] = nummax
ratelist[-1] = prediction[0][nummax]
poslist[-1] = iternum
else:
finallist[-1] = nummax
ratelist[-1] = prediction[0][nummax]
poslist[-1] = iternum
else:
finallist[-1] = nummax
ratelist[-1] = prediction[0][nummax]
poslist[-1] = iternum
#put into the list OR update, limit operation
elif (iternum - poslist[-1] <= 5 and iternum - poslist[-1] >= 4):
if (len(finallist) == 1): #add not update
if (nummax == 1):
if (prediction[0][nummax] > 0.999):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
elif (prediction[0][nummax] > 0.99):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
else:
if (ratelist[-1] > 0.999
and len(finallist) < 11): # last rate is too high, add
if (nummax == finallist[-1]
and finallist[-1] == finallist[-2]):
if (nummax != 1 and iternum - poslist[-2] > 14):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
elif (nummax == 1):
if (finallist[-1] != 0):
if (prediction[0][nummax] > 0.9999
and iternum - poslist[-2] > 14):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
elif (prediction[0][nummax] > 0.99):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
elif (ratelist[-1] > 0.98 and nummax == 9
and len(finallist) < 11): # 9 rate is lower, add
if (nummax == finallist[-1]
and finallist[-1] == finallist[-2]):
if (nummax != 1 and iternum - poslist[-2] > 14):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
elif (nummax == 1 and iternum - poslist[-2] > 14):
if (prediction[0][nummax] > 0.9999):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
elif (prediction[0][nummax] > 0.99):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
elif (prediction[0][nummax] > ratelist[-1]): #update
if (iternum - poslist[-2] < 14):
if (nummax == 1):
if (finallist[-1] == 0):
if (prediction[0][nummax] > 0.9999):
finallist[-1] = nummax
ratelist[-1] = prediction[0][nummax]
poslist[-1] = iternum
elif (prediction[0][nummax] > 0.999):
finallist[-1] = nummax
ratelist[-1] = prediction[0][nummax]
poslist[-1] = iternum
else:
finallist[-1] = nummax
ratelist[-1] = prediction[0][nummax]
poslist[-1] = iternum
#direct put into the list
elif ((iternum - poslist[-1]) > 5 and len(finallist) < 11):
if (len(finallist) <= 2):
if (nummax == 1):
if (prediction[0][nummax] > 0.993):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
else:
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
else: # 2 same num might detect as 3 nums
if (nummax == finallist[-1]
and finallist[-1] == finallist[-2]):
if (nummax != 1 and iternum - poslist[-2] > 14):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
else:
if (nummax == 1):
if (prediction[0][nummax] > 0.993):
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
else:
finallist.append(nummax)
ratelist.append(prediction[0][nummax])
poslist.append(iternum)
outfile.write(str(nummax) + ' ' + str(prediction[0][nummax]) + '\n')
#%timeit net.predict([input_image])
# Resize the image to the standard (256, 256) and oversample net input sized crops.
input_oversampled = caffe.io.oversample(
[caffe.io.resize_image(input_image, net.image_dims)],
net.crop_dims)
# 'data' is the input blob name in the model definition, so we preprocess for that input.
caffe_input = np.asarray(
[net.preprocess('data', in_) for in_ in input_oversampled])
# forward() takes keyword args for the input blobs with preprocessed input arrays.
#%timeit net.forward(data=caffe_input)
# print finallist
# print ratelist
# print poslist
# outfile.close()
outfile.write(str(finallist) + '\n')
outfile.write(str(ratelist) + '\n')
outfile.write(str(poslist) + '\n')
outfile.close()
str_out = ''
n_for = 0
while (n_for < len(finallist)):
str_out += str(finallist[n_for])
n_for = n_for + 1
# print str_out
return str(str_out)
def evaluate(self, metric='cosine'):
'''
@brief: 评测模型的性能
@param：itera： 模型的迭代次数
@param：metric： 度量的方法
'''
cap = cv2.VideoCapture(0)
caffe.set_mode_gpu()
net = caffe.Classifier(self.deploy,
self.model,
mean=np.load(self.imgmean_npy))
Lists = os.listdir(
'/home/kjin/caffe-master/examples/VGGNet/ContrastivePicNew/')
Feature = [None] * len(Lists)
i = 0
for list in Lists:
X1 = '/home/kjin/caffe-master/examples/VGGNet/ContrastivePicNew/' + list
X = DeepIDTest.read_imagelist(X1)
out = net.forward_all(data=X)
Feature[i] = np.float64(out['fc7'])
Feature[i] = np.reshape(Feature[i], 4096)
i = i + 1
prototxt = [
'/home/kjin/caffe-master/examples/VGGNet/pycaffe-mtcnn-master/model/'
+ x + '.prototxt' for x in ['det1', 'det2', 'det3']
]
binary = [
'/home/kjin/caffe-master/examples/VGGNet/pycaffe-mtcnn-master/model/'
+ x + '.caffemodel' for x in ['det1', 'det2', 'det3']
]
PNet = caffe.Net(prototxt[0], binary[0], caffe.TEST)
RNet = caffe.Net(prototxt[1], binary[1], caffe.TEST)
ONet = caffe.Net(prototxt[2], binary[2], caffe.TEST)
Error = 70
RecognitionNum = [None] * 5
while True:
while True:
t1 = time.time()
ret, im = cap.read()
# Load image.
# im = cv2.imread(frame)s
# assert im is not None, 'Image is empty.'
im_bk = im.copy()
im = im.astype(np.float32)
im = cv2.cvtColor(im, cv2.COLOR_BGR2RGB)
im = np.transpose(im, (1, 0, 2)) # Rotate image.
image_height, image_width, num_channels = im.shape
print('Image shape:', im.shape)
# assert num_channels == 3, 'Error: only support RGB image.'
MIN_FACE_SIZE = 24. # Minimum face size.
MIN_INPUT_SIZE = 12. # Minimum input size.
m = MIN_INPUT_SIZE / MIN_FACE_SIZE
min_size = min(image_height, image_width)
min_size = min_size * m
scales = []
counter = 0
FACTOR = 0.709
while min_size >= MIN_INPUT_SIZE:
scales.append(m * FACTOR**counter)
min_size = min_size * FACTOR
counter = counter + 1
# Load models.
# Threshold for each stage.
# THRESHOLD = [0.6, 0.7, 0.7]
THRESHOLD = [0.9, 0.9, 0.9]
#t1 = time.time()
# --------------------------------------------------------------
# First stage.
#
total_boxes = [] # Bounding boxes of all scales.
for scale in scales:
hs = int(math.ceil(image_height * scale))
ws = int(math.ceil(image_width * scale))
im_resized = cv2.resize(im, (ws, hs),
interpolation=cv2.INTER_AREA)
print('Resize to:', im_resized.shape)
# H,W,C -> C,H,W
im_resized = np.transpose(im_resized, (2, 0, 1))
# Zero mean and normalization.
im_resized = (im_resized - 127.5) * 0.0078125
# Reshape input layer.
PNet.blobs['data'].reshape(1, 3, hs, ws)
PNet.blobs['data'].data[...] = im_resized
outputs = PNet.forward()
bboxes = get_pnet_boxes(outputs, scale, THRESHOLD[0])
bboxes, _ = non_max_suppression(bboxes, 0.5)
total_boxes.append(bboxes)
total_boxes = np.vstack(total_boxes)
bboxes, _ = non_max_suppression(total_boxes, 0.7)
bboxes = bbox_regression(total_boxes)
bboxes = bbox_to_square(bboxes)
bboxes = padding(bboxes, image_height, image_width)
print('After PNet bboxes shape: ', bboxes.shape)
if bboxes.shape[0] == 0:
draw_and_show(im_bk, bboxes)
break
# --------------------------------------------------------------
# Second stage.
#
inputs = get_inputs_from_bboxes(im, bboxes, 24)
N, C, H, W = inputs.shape
RNet.blobs['data'].reshape(N, 3, H, W)
RNet.blobs['data'].data[...] = inputs
outputs = RNet.forward()
bboxes = get_rnet_boxes(bboxes, outputs, THRESHOLD[1])
bboxes, _ = non_max_suppression(bboxes, 0.7)
bboxes = bbox_regression(bboxes)
bboxes = bbox_to_square(bboxes)
bboxes = padding(bboxes, image_height, image_width)
print('After RNet bboxes shape: ', bboxes.shape)
if bboxes.shape[0] == 0:
draw_and_show(im_bk, bboxes)
break
# --------------------------------------------------------------
# Third stage.
#
inputs = get_inputs_from_bboxes(im, bboxes, 48)
N, C, H, W = inputs.shape
ONet.blobs['data'].reshape(N, 3, H, W)
ONet.blobs['data'].data[...] = inputs
outputs = ONet.forward()
bboxes, points = get_onet_boxes(bboxes, outputs, THRESHOLD[2])
bboxes = bbox_regression(bboxes)
bboxes, picked_indices = non_max_suppression(
bboxes, 0.7, 'min')
points = points[picked_indices]
bboxes = padding(bboxes, image_height, image_width)
print('After ONet bboxes shape: ', bboxes.shape, '\n')
if bboxes.shape[0] == 0:
draw_and_show(im_bk, bboxes, points)
break
# t2 = time.time()
#print('Total time: %.3fs\n' % (t2 - t1))
# draw_and_show(im_bk, bboxes, points)
num_boxes = bboxes.shape[0]
Boxnum = 0 #记录实际人脸个数
CoordinateList = [None] * num_boxes #记录每个人脸对应的坐标点
ROIimage = [None] * num_boxes
if num_boxes != 0:
for i in range(num_boxes):
box = bboxes[i]
x1 = int(box[0])
y1 = int(box[1])
x2 = int(box[2])
y2 = int(box[3])
if (x2 - x1 < 50) or (y2 - y1 < 50) or (y2 - y1 >
x2 - x1 - 30):
continue
cv2.rectangle(im_bk, (y1, x1), (y2, x2), (0, 255, 255),
2)
CoordinateList[Boxnum] = (y1, x1)
ROIimage[Boxnum] = im_bk[x1:x2 + 1, y1:y2 + 1]
Boxnum = Boxnum + 1
for i in range(Boxnum):
cv2.imwrite(
'/home/kjin/caffe-master/examples/VGGNet/' + str(i) +
'.jpg', ROIimage[i])
feature2 = [None] * Boxnum
for i in range(Boxnum):
X2 = '/home/kjin/caffe-master/examples/VGGNet/' + str(
i) + '.jpg'
#X2 = '/home/kjin/caffe-master/examples/VGGNet/pycaffe-mtcnn-master/Others/Repired/Test96.jpg'
X = DeepIDTest.read_imagelist(X2)
out = net.forward_all(data=X)
# print out
feature2[i] = np.float64(out['fc7'])
min_predicts = 1
for Num in range(Boxnum):
#t3=time.time()
# Calculate_Distance_1(Feature, feature2[Num], metric, min_predicts, Lists_Num)
# Calculate_Distance_2(Feature, feature2[Num], metric, min_predicts, Lists_Num)
Thread1 = threading.Thread(target=Calculate_Distance_1,
args=(Feature, feature2[Num],
metric, min_predicts,
len(Lists)))
Thread1.start()
#主线程运行，加上上面的线程，共２线程
global ThreadingState1
global ThreadingState2
ThreadingState1 = 0
ThreadingState2 = 0
i = 1
if len(Lists) / 2 is not (len(Lists) + 1) / 2: # 奇数
for sublist in range((len(Lists) - 1) / 2):
predicts2 = pw.pairwise_distances(Feature[i],
feature2[Num],
metric=metric)
i = i + 2
if predicts2[0][0] > 0.12:
if ThreadingState1 is 1:
break
if predicts2[0][0] < min_predicts:
min_predicts = predicts2[0][0]
else:
min_predicts = predicts2[0][0]
ThreadingState2 = 1
break
else:
for i in range(len(Lists) / 2): # 偶数
predicts2 = pw.pairwise_distances(Feature[i],
feature2[Num],
metric=metric)
i = i + 2
if predicts2[0][0] > 0.12:
if ThreadingState1 is 1:
break
if predicts2[0][0] < min_predicts:
min_predicts = predicts2[0][0]
else:
min_predicts = predicts2[0][0]
ThreadingState2 = 1
break
#Thread2=threading.Thread(target=Calculate_Distance_2,args=(Feature, feature2[Num], metric, min_predicts, len(Lists)))
#Thread2.start()
Thread1.join()
# Thread2.join()
# while ( Thread1.isAlive()) or ( Thread2.isAlive()):
# pass
#print ('Total time: %.3fs\n' % (t2 - t1))
#predicts = pw.pairwise_distances(Feature[i], feature2[Num], metric=metric)
# i=i+1
# predicts = mt
# if predicts[0][0] >0.12:
# if predicts[0][0] < min_predicts:
# min_predicts = predicts[0][0]
# if i==(len(Lists)-1):
# accuracy=1
if (ThreadingState1 == 0) and (ThreadingState2 == 0):
if RecognitionNum[Num] is None:
RecognitionNum[Num] = 1
else:
RecognitionNum[Num] = RecognitionNum[Num] + 1
if RecognitionNum[Num] > 0:
cv2.putText(im_bk, 'Unknows', CoordinateList[Num],
cv2.FONT_HERSHEY_COMPLEX_SMALL, 2,
(0, 0, 255))
#accuracy=1
#cv2.imwrite('/home/kjin/caffe-master/examples/VGGNet/ContrastivePicNew/AddError_'+str(Error)+'.jpg', ROIimage[Num])
# Error=Error+1
RecognitionNum[Num] = 0
# t3=time.time()
# os.system('sh /home/kjin/caffe-master/examples/VGGNet/beep.sh')
# t4=time.time()
#print('Beep time: %.3fs\n' % (t4-t3))
elif ThreadingState1 or ThreadingState2:
RecognitionNum[Num] = 0
# min_predicts=predicts[0][0]
#accuracy=0
cv2.putText(im_bk, 'Me', CoordinateList[Num],
cv2.FONT_HERSHEY_COMPLEX_SMALL, 2,
(0, 255, 0))
# break #若是本人，立马break，不用再比较
cv2.imshow('result', im_bk)
t2 = time.time()
cv2.waitKey(8)
print('Total time: %.3fs\n' % (t2 - t1))
if SUBMIT:
TEST_DB = "plankton_test_lmdb"
FEATURES_FILE = "test_features.npz"
else:
# TEST_DB='plankton_val_lmdb'
TEST_DB = 'plankton_train_lmdb'
FEATURES_FILE = "val_features.npz"
ENCODE_FILE = "/home/shai/mocha/data/plankton/encode.txt"
MODEL_FILE = 'inet_deploy8.prototxt'
PRETRAINED = 'inet9.caffemodel'
print "Try to create net..."
net = caffe.Classifier(
MODEL_FILE,
PRETRAINED,
# mean=np.load('mean.npy'),
image_dims=(tsize, tsize))
print "Created net"
net.set_phase_test()
net.set_mode_gpu()
env = lmdb.open(TEST_DB)
datum = caffe.proto.caffe_pb2.Datum()
def visualize(features):
model = TSNE(n_components=2, random_state=0)
X = model.fit_transform(features)
print X.shape
fig, ax = plt.subplots()
proto='./gdep.prototxt'
caffe.set_mode_gpu()
bsize=512
mean = np.zeros((3,int(bsize),int(bsize)))
mean[0,:,:]=104.00699
mean[1,:,:]=116.66877
mean[2,:,:]=122.67892
channel_swap = [2,1,0]
raw_scale=255.0;
center_only=False
input_scale=None
image_dims=[bsize,bsize]
ims=(int(bsize),int(bsize))
data = caffe.Classifier(proto,cm, image_dims=image_dims,mean=mean,
input_scale=input_scale,
raw_scale=raw_scale,
channel_swap=channel_swap)
im = [caffe.io.load_image('./img.jpg')]
im2=[caffe.io.resize_image(im[0], ims)]
im3 = np.zeros((1,ims[0],ims[1],im2[0].shape[2]),dtype=np.float32)
im3[0]=im2[0]
caffe_in = np.zeros(np.array(im3.shape)[[0, 3, 1, 2]],dtype=np.float32)
caffe_in[0]=data.transformer.preprocess('data', im3[0])
out = data.forward_all(**{'data': caffe_in})
map=data.blobs['dcsmn'].data
id=data.blobs['sortid'].data
rsize=512
zeronp=np.zeros((rsize,rsize),dtype=np.float32)
ctg=['plane','bicycle','bird','boat','bottle','bus','car','cat','chair','cow','dog','dtable','horse','moterbike','person','plant','sheep','sofa','train','tv']
for i in range(map.shape[0]):
def main():
parser = argparse.ArgumentParser(description='Process some integers.')
parser.add_argument('--unit', metavar='unit', type=int, help='an unit to visualize e.g. [0, 999]')
parser.add_argument('--n_iters', metavar='iter', type=int, default=10, help='Number of iterations')
parser.add_argument('--L2', metavar='w', type=float, default=1.0, nargs='?', help='L2 weight')
parser.add_argument('--start_lr', metavar='lr', type=float, default=2.0, nargs='?', help='Learning rate')
parser.add_argument('--end_lr', metavar='lr', type=float, default=-1.0, nargs='?', help='Ending Learning rate')
parser.add_argument('--seed', metavar='n', type=int, default=0, nargs='?', help='Learning rate')
parser.add_argument('--xy', metavar='n', type=int, default=0, nargs='?', help='Spatial position for conv units')
parser.add_argument('--opt_layer', metavar='s', type=str, help='Layer at which we optimize a code')
parser.add_argument('--act_layer', metavar='s', type=str, default="fc8", help='Layer at which we activate a neuron')
parser.add_argument('--init_file', metavar='s', type=str, default="None", help='Init image')
parser.add_argument('--debug', metavar='b', type=int, default=0, help='Print out the images or not')
parser.add_argument('--clip', metavar='b', type=int, default=0, help='Clip out within a code range')
parser.add_argument('--bound', metavar='b', type=str, default="", help='The file to an array that is the upper bound for activation range')
parser.add_argument('--output_dir', metavar='b', type=str, default=".", help='Output directory for saving results')
parser.add_argument('--net_weights', metavar='b', type=str, default=settings.net_weights, help='Weights of the net being visualized')
parser.add_argument('--net_definition', metavar='b', type=str, default=settings.net_definition, help='Definition of the net being visualized')
parser.add_argument('--outName' , metavar='s',type=str,help='name of the output file')
parser.add_argument('--appendImages', action='append')
parser.add_argument('--appendNames', action='append')
args = parser.parse_args()
# Default to constant learning rate
if args.end_lr < 0:
args.end_lr = args.start_lr
# which neuron to visualize
print "-------------"
print " unit: %s xy: %s" % (args.unit, args.xy)
print " n_iters: %s" % args.n_iters
print " L2: %s" % args.L2
print " start learning rate: %s" % args.start_lr
print " end learning rate: %s" % args.end_lr
print " seed: %s" % args.seed
print " opt_layer: %s" % args.opt_layer
print " act_layer: %s" % args.act_layer
print " init_file: %s" % args.init_file
print " clip: %s" % args.clip
print " bound: %s" % args.bound
print "-------------"
print " debug: %s" % args.debug
print " output dir: %s" % args.output_dir
print " net weights: %s" % args.net_weights
print " net definition: %s" % args.net_definition
print "-------------"
params = [
{
'layer': args.act_layer,
'iter_n': args.n_iters,
'L2': args.L2,
'start_step_size': args.start_lr,
'end_step_size': args.end_lr
}
]
act_unit = args.unit
caffe.set_mode_gpu()
#caffe.set_mode_cpu()
# networks
generator = caffe.Net(settings.generator_definition, settings.generator_weights, caffe.TEST)
net = caffe.Classifier(args.net_definition, args.net_weights,
mean = mean, # ImageNet mean
channel_swap = (2,1,0)) # the reference model has channels in BGR order instead of RGB
# input / output layers in generator
gen_in_layer = "feat"
gen_out_layer = "deconv0"
# shape of the code being optimized
shape = generator.blobs[gen_in_layer].data.shape
# Fix the seed
np.random.seed(args.seed)
image_list = args.appendImages
#print image_list
name_list = args.appendNames
#print name_list
# Optimize a code via gradient ascent
for i in range(len(name_list)):
start = time.time()
if len(image_list) > 0:
start_code, start_image = get_code(image_list[i], args.opt_layer)
print "Loaded start code: ", start_code.shape
else:
start_code = np.random.normal(0, 1, shape)
# Load the activation range
upper_bound = lower_bound = None
# Set up clipping bounds
if args.bound != "":
n_units = shape[1]
upper_bound = np.loadtxt(args.bound, delimiter=' ', usecols=np.arange(0, n_units), unpack=True)
upper_bound = upper_bound.reshape(start_code.shape)
# Lower bound of 0 due to ReLU
lower_bound = np.zeros(start_code.shape)
print "Running file %d or %d , filename : %s"%(i , len(name_list) , image_list[i] )
output_image , best_act_final = activation_maximization(net, generator, gen_in_layer, gen_out_layer, start_code, params,
clip=args.clip, unit=act_unit, xy=args.xy, debug=args.debug,
upper_bound=upper_bound, lower_bound=lower_bound)
print output_image.shape
#Save image
#filename = "%s/%s_%s_%s_%s_%s_%s__%s.jpg" % (
# args.output_dir,
# args.init_file.split('.')[0],
# args.act_layer,
# str(args.unit).zfill(4),
# str(args.n_iters).zfill(2),
# args.L2,
# args.start_lr,
# args.seed
# )
end = time.time()
print '%30s' % 'Executed DGN-AM in ', str((end - start)*1000), 'ms'
start = time.time()
filename = "%s/%s_%s_%s_.jpg" % (
args.output_dir,
name_list[i],
str(args.unit).zfill(4),
str(best_act_final)
)
# Save image
save_image(output_image, filename)
print "Saved to %s" % filename
end = time.time()
print '%30s' % 'Saved File in in ', str((end - start)*1000), 'ms'
def main(
input,
output,
image_type,
gpu,
model_path,
model_name,
preview,
octaves,
octave_scale,
iterations,
jitter,
zoom,
stepsize,
blend,
layers,
guide_image,
start_frame,
end_frame,
verbose,
):
make_sure_path_exists(input)
make_sure_path_exists(output)
# let max nr of frames
nrframes = len([
name for name in os.listdir(input)
if os.path.isfile(os.path.join(input, name))
])
if nrframes == 0:
print('no frames to process found')
sys.exit(0)
if preview is None:
preview = 0
if octaves is None:
octaves = 4
if octave_scale is None:
octave_scale = 1.5
if iterations is None:
iterations = 5
if jitter is None:
jitter = 32
if zoom is None:
zoom = 1
if stepsize is None:
stepsize = 1.5
if blend is None: # can be nr (constant), random, or loop
blend = 0.5
if verbose is None:
verbose = 1
if layers is None: # ['inception_4c/output']
layers = 'customloop'
if start_frame is None:
frame_i = 1
else:
frame_i = int(start_frame)
if not end_frame is None:
nrframes = int(end_frame) + 1
else:
nrframes = nrframes + 1
# Load DNN
net_fn = model_path + 'deploy.prototxt'
param_fn = model_path + model_name # 'bvlc_googlenet.caffemodel'
if gpu is None:
print("SHITTTTTTTTTTTTTT You're running CPU man =D")
else:
caffe.set_mode_gpu()
caffe.set_device(int(args.gpu))
print('GPU mode [device id: %s]' % args.gpu)
print("using GPU, but you'd still better make a cup of coffee")
# Patching model to be able to compute gradients.
# Note that you can also manually add "force_backward: true" line to "deploy.prototxt".
model = caffe.io.caffe_pb2.NetParameter()
text_format.Merge(open(net_fn).read(), model)
model.force_backward = True
open('tmp.prototxt', 'w').write(str(model))
net = caffe.Classifier(
'tmp.prototxt',
param_fn,
mean=np.float32([104.0, 116.0, 122.0]),
channel_swap=(2, 1, 0)) # ImageNet mean, training set dependent
# the reference model has channels in BGR order instead of RGB
if verbose == 3:
from IPython.display import clear_output, Image, display
print('display turned on')
frame = np.float32(
PIL.Image.open(input + '/%08d.%s' % (frame_i, image_type)))
if preview is not 0:
frame = np.float32(
resizePicture(input + '/%08d.%s' % (frame_i, image_type), preview))
now = time.time()
totaltime = 0
if blend == 'loop':
blend_forward = True
blend_at = 0.4
blend_step = 0.1
for i in range(frame_i, nrframes):
print('Processing frame #{}'.format(frame_i))
# Choosing Layer
if layers == 'customloop': # loop over layers as set in layersloop array
endparam = layersloop[frame_i % len(layersloop)]
else:
# loop through layers one at a time until this specific layer
endparam = layers[frame_i % len(layers)]
# Choosing between normal dreaming, and guided dreaming
if guide_image is None:
frame = deepdream(
net,
frame,
image_type=image_type,
verbose=verbose,
iter_n=iterations,
step_size=stepsize,
octave_n=octaves,
octave_scale=octave_scale,
jitter=jitter,
end=endparam,
)
else:
guide = np.float32(PIL.Image.open(guide_image))
print('Setting up Guide with selected image')
guide_features = prepare_guide(net,
PIL.Image.open(guide_image),
end=endparam)
frame = deepdream_guided(
net,
frame,
image_type=image_type,
verbose=verbose,
iter_n=iterations,
step_size=stepsize,
octave_n=octaves,
octave_scale=octave_scale,
jitter=jitter,
end=endparam,
objective_fn=objective_guide,
guide_features=guide_features,
)
saveframe = output + '/%08d.%s' % (frame_i, image_type)
later = time.time()
difference = int(later - now)
totaltime += difference
avgtime = totaltime / i
# Stats (stolen + adapted from Samim: https://github.com/samim23/DeepDreamAnim/blob/master/dreamer.py)
print('***************************************')
print('Saving Image As: ' + saveframe)
print('Frame ' + str(i) + ' of ' + str(nrframes - 1))
print('Frame Time: ' + str(difference) + 's')
timeleft = avgtime * (nrframes - 1 - frame_i)
(m, s) = divmod(timeleft, 60)
(h, m) = divmod(m, 60)
print('Estimated Total Time Remaining: ' + str(timeleft) \
+ 's (' + '%d:%02d:%02d' % (h, m, s) + ')')
print('***************************************')
PIL.Image.fromarray(np.uint8(frame)).save(saveframe)
newframe = input + '/%08d.%s' % (frame_i, image_type)
if blend == 0:
newimg = PIL.Image.open(newframe)
if preview is not 0:
newimg = resizePicture(newframe, preview)
frame = newimg
else:
if blend == 'random':
blendval = randint(5, 10) / 10.
elif blend == 'loop':
if blend_at > 1 - blend_step:
blend_forward = False
elif blend_at <= 0.5:
blend_forward = True
if blend_forward:
blend_at += blend_step
else:
blend_at -= blend_step
blendval = blend_at
else:
blendval = float(blend)
frame = morphPicture(saveframe, newframe, blendval, preview)
frame = np.float32(frame)
now = time.time()
frame_i += 1
def main():
inputfile = '/media/jwong/Transcend/VQADataset/imagePaths.txt'
outputfile = '/media/jwong/Transcend/VQADataset/VQATrainOutput'
jsonFile = '/media/jwong/Transcend/VQADataset/VQAImgFeatures_Train.json'
errorLogFile = '/media/jwong/Transcend/VQADataset/ImgFeatures_TrainLog.txt'
print 'Reading images from "', inputfile
print 'Writing vectors to "', outputfile
print 'Writing to json file "', jsonFile
# Setting this to CPU, but feel free to use GPU if you have CUDA installed
caffe.set_mode_cpu()
# Loading the Caffe model, setting preprocessing parameters
net = caffe.Classifier(model_prototxt,
model_trained,
mean=np.load(mean_path).mean(1).mean(1),
channel_swap=(2, 1, 0),
raw_scale=255,
image_dims=(480, 640))
# Loading class labels
with open(imagenet_labels) as f:
labels = f.readlines()
#Names & sizes of network layers
#print [(k, v.data.shape) for k, v in net.blobs.items()]
resultJSONData = {}
errorMessages = []
count = 0
# Processing one image at a time, print predictions
with open(inputfile, 'r') as reader:
with open(outputfile, 'w') as writer:
writer.truncate()
for image_path in reader:
image_path = image_path.strip()
input_image = caffe.io.load_image(image_path)
prediction = net.predict([input_image], oversample=False)
print os.path.basename(image_path), ' : ', labels[
prediction[0].argmax()].strip(), ' (', prediction[0][
prediction[0].argmax()], ')'
count = count + 1
try:
#Get image ID
splitPath = image_path.split('/')
imgNameParts = splitPath[len(splitPath) - 1].split(
'_') #COCO, train, XXX.jpg
suffix = imgNameParts[len(imgNameParts) - 1] #XXX.jpg
img_id = int(suffix.split('.')[0])
print 'Reading img ', img_id
print 'Images processed : ', count
# filename, array data to be saved, format, delimiter
featureData = net.blobs[layer_name].data[0].reshape(
1, -1).tolist()
np.savetxt(writer, featureData, fmt='%.8g')
resultJSONData[img_id] = featureData
except ValueError:
#Invalid image names
errorMessages.append(image_path)
with open(jsonFile, 'w') as jsonOut:
json.dump(resultJSONData, jsonOut)
with open(errorLogFile, 'w') as logFile:
for msg in errorMessages:
logFile.write(msg)
def InitializeDetecter(self):
self._net = caffe.Classifier(self._NET_FILE, self._TRAINED_MODEL_FILE)
if os.path.exists(self._outputDir) is False:
os.system('mkdir ' + self._outputDir)
import pickle
import skimage
import redis
caffe_root = '/home/jiangwei/bin/caffe-master/'
import sys
sys.path.insert(0, caffe_root + 'python')
import caffe
import sklearn.metrics.pairwise as pw
#GPU加速
caffe.set_mode_gpu()
#加载caffe模型
global net
net=caffe.Classifier('vgg/deploy.prototxt','vgg/vgg_face.caffemodel',caffe.TEST)
#提取特征数组
def get_feature(path1):
global net
#加载图片
X=read_image(path1)
test_num=np.shape(X)[0]
#X 作为 模型的输入
out = net.forward_all(blobs=['pool5'],data = X)
# print out.keys()
feature1 = np.float64(out["pool5"])
#load our mean file and reshape it accordingly
a = caffe.io.caffe_pb2.BlobProto()
file = open('%s/DB_train_w32_%d.binaryproto' % (BASE, FOLD), 'rb')
data = file.read()
a.ParseFromString(data)
means = a.data
means = np.asarray(means)
means = means.reshape(3, 32, 32)
#make sure we use teh GPU otherwise things will take a very long time
caffe.set_mode_gpu()
#load the model
net = caffe.Classifier(MODEL_FILE,
PRETRAINED,
mean=means,
channel_swap=(2, 1, 0),
raw_scale=255,
image_dims=(32, 32))
#see which files we need to produce output for in this fold
#we look at the parent IDs in the test file and only compute those images
#as they've been "held out" from the training set
files = open('%s/test_w32_parent_%d.txt' % (BASE, FOLD), 'rb')
start_time = time.time()
start_time_iter = 0
#go into the image directory so we can use glob a bit easier
os.chdir(IMAGE_DIR)
for base_fname in files:
import numpy as np
import numpy as np
import os
import skimage
import sys
import caffe
import math
import sklearn.metrics.pairwise as pw
from fr_wuqianliang_sphereface import *
from matlab_cp2tform import get_similarity_transform_for_cv2
# load Caffe model
global net
net = caffe.Classifier('sphereface_deploy.prototxt',
'sphereface_model_iter_28000.caffemodel')
def getface(im, boxes):
constImgfile = "./tmp"
for i in range(len(boxes)):
cv2.rectangle(im, (int(boxes[i][0]), int(boxes[i][1])),
(int(boxes[i][2]), int(boxes[i][3])), (0, 255, 0), 1)
cropped = im[int(boxes[i][1]):int(boxes[i][3]),
int(boxes[i][0]):int(boxes[i][2]), :]
#cropped = cv2.resize(cropped, (96, 96), interpolation=cv2.INTER_CUBIC )
img = cropped
#cv2.imshow('src',img)
# File containing the class labels
imagenet_labels = caffe_root + 'data/ilsvrc12/synset_words.txt'
# Path to the mean image (used for input processing)
mean_path = caffe_root + 'python/caffe/imagenet/ilsvrc_2012_mean.npy'
# Name of the layer we want to extract
layer_name = 'pool5/7x7_s1'
sys.path.insert(0, caffe_root + 'python')
import caffe
caffe.set_mode_gpu()
net = caffe.Classifier(model_prototxt,
model_trained,
mean=np.load(mean_path).mean(1).mean(1),
channel_swap=(2, 1, 0),
raw_scale=255,
image_dims=(256, 256))
# print [(k, v.data.shape) for k, v in net.blobs.items()]
# Loading class labels
with open(imagenet_labels) as f:
labels = f.readlines()
def forward_cnn(image_path):
image_path = image_path.strip()
input_image = caffe.io.load_image(image_path)
prediction = net.predict([input_image], oversample=False)
print os.path.basename(image_path), ' : ', labels[prediction[0].argmax(
)].strip(), ' (', prediction[0][prediction[0].argmax()], ')'
import sys
import caffe
import matplotlib.pyplot as plt
import os
caffe_root = '../'
MODEL_FILE = 'mnist/lenet.prototxt'
PRETRAINED = 'mnist/lenet/lenet_iter_10000.caffemodel'
IMAGE_FILE = 'mnist/slike/test7-4.png' #image path
input_image = caffe.io.load_image(IMAGE_FILE, color=False)
digits = [0, 0, 0, 0, 0, 0, 0, 0, 0, 0]
for i in range(50):
net = caffe.Classifier(
MODEL_FILE, PRETRAINED, raw_scale=255,
image_dims=(28,
28)) #, image_dims=(28, 28), raw_scale=0.00392156862745
caffe.set_mode_cpu()
score = net.predict([input_image], oversample=False)
if np.argmax(score) == 0:
digits[0] += 1
continue
elif np.argmax(score) == 1:
digits[1] += 1
continue
elif np.argmax(score) == 2:
digits[2] += 1
continue
elif np.argmax(score) == 3:
digits[3] += 1
continue
caffe_root = './caffe-master/'
sys.path.insert(0, caffe_root + 'python')
sys.path.append('/home/gong/Downloads/caffe-master/python')
import caffe
mean_filename='./mean.binaryproto'
proto_data = open(mean_filename, "rb").read()
a = caffe.io.caffe_pb2.BlobProto.FromString(proto_data)
mean = caffe.io.blobproto_to_array(a)[0]
gender_net_pretrained='./gender_train_net.caffemodel'
gender_net_model_file='./deploy_gender.prototxt'
gender_net = caffe.Classifier(gender_net_model_file, gender_net_pretrained,
mean=mean,
channel_swap=(2,1,0),
raw_scale=127,
image_dims=(128, 128))
gender_list=['Male','Female']
## ===============read the images in folder './image' ==============##
#path = './image'
#images = os.listdir(path)
#n = 1;
#for file in images:
# if not os.path.isdir(file):
# example_image = open(path + "/" + file)
# input_image = caffe.io.load_image(example_image)
# _ = plt.imshow(input_image)
def main():
parser = argparse.ArgumentParser(description='Process some integers.')
parser.add_argument('--units',
metavar='units',
type=str,
help='an unit to visualize e.g. [0, 999]')
parser.add_argument('--n_iters',
metavar='iter',
type=int,
default=10,
help='Number of sampling steps per each unit')
parser.add_argument(
'--threshold',
metavar='w',
type=float,
default=-1.0,
nargs='?',
help='The probability threshold to decide whether to keep an image')
parser.add_argument(
'--save_every',
metavar='save_iter',
type=int,
default=1,
help='Save a sample every N iterations. 0 to disable saving')
parser.add_argument('--reset_every',
metavar='reset_iter',
type=int,
default=0,
help='Reset the code every N iterations')
parser.add_argument('--lr',
metavar='lr',
type=float,
default=2.0,
nargs='?',
help='Learning rate')
parser.add_argument('--lr_end',
metavar='lr',
type=float,
default=-1.0,
nargs='?',
help='Ending Learning rate')
parser.add_argument('--epsilon2',
metavar='lr',
type=float,
default=1.0,
nargs='?',
help='Ending Learning rate')
parser.add_argument('--epsilon1',
metavar='lr',
type=float,
default=1.0,
nargs='?',
help='Ending Learning rate')
parser.add_argument('--epsilon3',
metavar='lr',
type=float,
default=1.0,
nargs='?',
help='Ending Learning rate')
parser.add_argument('--seed',
metavar='n',
type=int,
default=0,
nargs='?',
help='Random seed')
parser.add_argument('--xy',
metavar='n',
type=int,
default=0,
nargs='?',
help='Spatial position for conv units')
parser.add_argument('--opt_layer',
metavar='s',
type=str,
help='Layer at which we optimize a code')
parser.add_argument('--act_layer',
metavar='s',
type=str,
default="fc8",
help='Layer at which we activate a neuron')
parser.add_argument('--init_file',
metavar='s',
type=str,
default="None",
help='Init image')
parser.add_argument('--write_labels',
action='store_true',
default=False,
help='Write class labels to images')
parser.add_argument('--output_dir',
metavar='b',
type=str,
default=".",
help='Output directory for saving results')
parser.add_argument('--net_weights',
metavar='b',
type=str,
default=settings.encoder_weights,
help='Weights of the net being visualized')
parser.add_argument('--net_definition',
metavar='b',
type=str,
default=settings.encoder_definition,
help='Definition of the net being visualized')
args = parser.parse_args()
# Default to constant learning rate
if args.lr_end < 0:
args.lr_end = args.lr
# summary
print "-------------"
print " units: %s xy: %s" % (args.units, args.xy)
print " n_iters: %s" % args.n_iters
print " reset_every: %s" % args.reset_every
print " save_every: %s" % args.save_every
print " threshold: %s" % args.threshold
print " epsilon1: %s" % args.epsilon1
print " epsilon2: %s" % args.epsilon2
print " epsilon3: %s" % args.epsilon3
print " start learning rate: %s" % args.lr
print " end learning rate: %s" % args.lr_end
print " seed: %s" % args.seed
print " opt_layer: %s" % args.opt_layer
print " act_layer: %s" % args.act_layer
print " init_file: %s" % args.init_file
print "-------------"
print " output dir: %s" % args.output_dir
print " net weights: %s" % args.net_weights
print " net definition: %s" % args.net_definition
print "-------------"
# encoder and generator for images
encoder = caffe.Net(settings.encoder_definition, settings.encoder_weights,
caffe.TEST)
generator = caffe.Net(settings.generator_definition,
settings.generator_weights, caffe.TEST)
# condition network, here an image classification net
net = caffe.Classifier(
args.net_definition,
args.net_weights,
mean=np.float32([104.0, 117.0, 123.0]), # ImageNet mean
channel_swap=(
2, 1,
0)) # the reference model has channels in BGR order instead of RGB
# Fix the seed
np.random.seed(args.seed)
if args.init_file != "None":
start_code, start_image = get_code(encoder=encoder,
path=args.init_file,
layer=args.opt_layer)
print "Loaded start code: ", start_code.shape
else:
# shape of the code being optimized
shape = generator.blobs[settings.generator_in_layer].data.shape
start_code = np.random.normal(0, 1, shape)
print ">>", np.min(start_code), np.max(start_code)
# Separate the dash-separated list of units into numbers
conditions = [{
"unit": int(u),
"xy": args.xy
} for u in args.units.split("_")]
# Optimize a code via gradient ascent
sampler = ClassConditionalSampler()
output_image, list_samples = sampler.sampling(
condition_net=net,
image_encoder=encoder,
image_generator=generator,
gen_in_layer=settings.generator_in_layer,
gen_out_layer=settings.generator_out_layer,
start_code=start_code,
n_iters=args.n_iters,
lr=args.lr,
lr_end=args.lr_end,
threshold=args.threshold,
layer=args.act_layer,
conditions=conditions,
epsilon1=args.epsilon1,
epsilon2=args.epsilon2,
epsilon3=args.epsilon3,
output_dir=args.output_dir,
reset_every=args.reset_every,
save_every=args.save_every)
# Output image
filename = "%s/%s_%04d_%04d_%s_h_%s_%s_%s__%s.jpg" % (
args.output_dir, args.act_layer, conditions[0]["unit"], args.n_iters,
args.lr, str(args.epsilon1), str(args.epsilon2), str(
args.epsilon3), args.seed)
# Save the final image
util.save_image(output_image, filename)
print "%s/%s" % (os.getcwd(), filename)
# Write labels to images
print "Saving images..."
for p in list_samples:
img, name, label = p
util.save_image(img, name)
if args.write_labels:
util.write_label_to_img(name, label)
import pandas as pd
import numpy as np
import caffe
import os
import glob
import re
net = caffe.Classifier('/home/seiji/caffe/swagnet_deploy.prototxt',
'/home/seiji/Desktop/swaga_iter_50000.caffemodel',
mean=np.load('/home/seiji/out.npy'),
channel_swap=(2, 1, 0),
raw_scale=255,
image_dims=(512, 512))
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2, 0, 1))
results = pd.DataFrame(columns=['filename', 'prediction'])
net.blobs['data'].reshape(1, 3, 512, 512)
n = 0
path = '/home/seiji/Desktop/Data/predictions/test/'
listing1 = os.listdir(path)
listing = sorted(listing1, key=lambda x: (int(re.sub('\D', '', x)), x))
for infile in listing:
input_image = caffe.io.load_image(
'/home/seiji/Desktop/Data/predictions/test/' + infile)
prediction = net.predict([input_image], oversample=False)
print 'predicted class:', prediction
pred = prediction.argmax()
results.loc[n] = np.array([infile, pred])
n += 1
def get_codes_and_labels(params):
device_id = int(params["gpu"])
caffe.set_device(device_id)
caffe.set_mode_gpu()
model_file = params['model_file']
pretrained_model = params['pretrained_model']
dims = params['image_dims']
scale = params['scale']
database = params['database']
validation = params['validation']
batch_size = params['batch_size']
net = caffe.Classifier(model_file,
pretrained_model,
channel_swap=(2, 1, 0),
image_dims=dims,
mean=np.array([103.939, 116.779, 123.68]),
raw_scale=scale)
database_code = []
validation_code = []
database_labels = []
validation_labels = []
cur_pos = 0
while 1:
lines = database[cur_pos:cur_pos + batch_size]
if len(lines) == 0:
break
cur_pos = cur_pos + len(lines)
images = [
caffe.io.load_image(line.strip().split(" ")[0]) for line in lines
]
labels = [[int(i) for i in line.strip().split(" ")[1:]]
for line in lines]
codes = net.predict(images, oversample=False)
#codes = net.predict(images)
[database_code.append(c) for c in codes]
[database_labels.append(l) for l in labels]
print str(cur_pos) + "/" + str(len(database))
if len(lines) < batch_size:
break
cur_pos = 0
while 1:
lines = validation[cur_pos:cur_pos + batch_size]
if len(lines) == 0:
break
cur_pos = cur_pos + len(lines)
images = [
caffe.io.load_image(line.strip().split(" ")[0]) for line in lines
]
labels = [[int(i) for i in line.strip().split(" ")[1:]]
for line in lines]
codes = net.predict(images, oversample=False)
#codes = net.predict(images)
[validation_code.append(c) for c in codes]
[validation_labels.append(l) for l in labels]
print str(cur_pos) + "/" + str(len(validation))
if len(lines) < batch_size:
break
return dict(database_code=np.sign(np.array(database_code)),
database_labels=np.array(database_labels),
validation_code=np.sign(np.array(validation_code)),
validation_labels=np.array(validation_labels))
# PRETRAINED = '../examples/kaggle2/googlenet/bvlc_googlenet_lr_1e-3_layer_2_iter_125000.caffemodel'
# PRETRAINED = '../examples/kaggle2/googlenet/bvlc_googlenet_txt_lr_1e-2_1020_iter_155000.caffemodel'
# PRETRAINED = '../examples/kaggle2/googlenet/bvlc_googlenet_txt_lr_1e-3_1020_iter_95000.caffemodel'
# PRETRAINED = '../examples/kaggle2/googlenet/bvlc_googlenet_lr_1e-3_layer_2_iter_155000.caffemodel'
# PRETRAINED = '../examples/kaggle2/googlenet/bvlc_googlenet_txt_lr_1e-2_1020_iter_160000.caffemodel'
# PRETRAINED = '../examples/kaggle2/googlenet/bvlc_googlenet_lr_1e-3_layer_2_iter_190000.caffemodel'
# PRETRAINED = '../examples/kaggle2/googlenet/bvlc_googlenet_txt_lr_1e-2_1020_iter_200000.caffemodel'
PRETRAINED = '/media/VSlab/coldmanck/fast-rcnn/caffe-fast-rcnn/examples/kaggle2/googlenet/bvlc_googlenet_txt_lr_1e-2_resize_rotated_iter_15000.caffemodel'
caffe.set_device(6)
caffe.set_mode_gpu()
net = caffe.Classifier(
MODEL_FILE,
PRETRAINED,
mean=np.load('/home/coldmanck/kaggle/kaggle_mean.npy').mean(1).mean(1),
# mean=np.load('../python/caffe/imagenet/ilsvrc_2012_mean.npy').mean(1).mean(1),
channel_swap=(2, 1, 0),
raw_scale=255,
image_dims=(224, 224))
total_num = 0
correct = 0
correct5 = 0
'''
net = caffe.Net(MODEL_FILE, PRETRAINED, caffe.TEST)
# input preprocessing: 'data' is the name of the input blob == net.inputs[0]
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_transpose('data', (2,0,1))
transformer.set_mean('data', np.load('/home/coldmanck/kaggle/kaggle_mean.npy').mean(1).mean(1)) # mean pixel
transformer.set_raw_scale('data', 255) # the reference model operates on images in [0,255] range instead of [0,1]
# -*- coding: utf-8 -*-
import os
import sys
import numpy as np
import matplotlib.pyplot as plt
caffe_root = os.environ.get('CAFFE_ROOT')
sys.path.insert(0, caffe_root + '/python') #把pycaffe所在路径添加到环境变量
import caffe
#指定网络结构 与 lenet_train_test.prototxt不同
MODEL_FILE = 'lenet.prototxt'
PRETRAINED = 'quick_draw_lenet_iter_10000.caffemodel'
#图片已经处理成 lenet.prototxt的输入要求（尺寸28x28）且已经二值化为黑白色
IMAGE_FILE = '../data/quickdraw_val_image/full_Fnumpy_bitmap_Fcat_6501.jpg'
# print os.path.exists(MODEL_FILE)
# print os.path.exists(PRETRAINED)
# print os.path.exists(IMAGE_FILE)
input_image = caffe.io.load_image(IMAGE_FILE, color=False)
net = caffe.Classifier(MODEL_FILE, PRETRAINED)
prediction = net.predict([input_image], oversample=False)
caffe.set_mode_cpu()
print 'predicted class:', prediction
import sys
sys.path.insert(0, caffe_root + 'python')
import caffe
# DEFN
curdir, _ = os.path.split(__file__)
MODEL_FILE = os.path.join(curdir, '../deploy_lb.prototxt')
PRETRAINED = os.path.join(curdir, '../models/lb65_iter_100000.caffemodel')
MEAN_FILE = os.path.join(curdir, '../data/64x64/ndsb_mean_test.npy')
TEST_FILE = os.path.join(curdir, '../data/test_final.txt')
N = 20000 # batch size for our pythno script to feed to caffe to avoid memory issues
net = caffe.Classifier(MODEL_FILE,
PRETRAINED,
mean=np.load(MEAN_FILE),
raw_scale=255,
image_dims=(57, 57),
gpu=True)
caffe.set_mode_gpu()
caffe.set_phase_test()
# Loading test paths
test_doc = np.genfromtxt(TEST_FILE, dtype='str')
test_files = test_doc
# Partition test_files for memory sake
def chunks(l, n):
""" Yield successive n-sized chunks from l.
"""
#for i in xrange(0, len(l), n):
def main(argv):
pycaffe_dir = os.path.dirname(__file__)
parser = argparse.ArgumentParser()
print "here"
# Required arguments: input and output files.
parser.add_argument("input_file", help="Input image, directory, or npy.")
parser.add_argument("output_file", help="Output npy filename.")
# Optional arguments.
parser.add_argument(
"--model_def",
default=os.path.join(
pycaffe_dir, "../models/bvlc_reference_caffenet/deploy.prototxt"),
help="Model definition file.")
parser.add_argument(
"--pretrained_model",
default=os.path.join(
pycaffe_dir,
"../models/bvlc_reference_caffenet/bvlc_reference_caffenet.caffemodel"
),
help="Trained model weights file.")
parser.add_argument("--gpu",
action='store_true',
help="Switch for gpu computation.")
parser.add_argument(
"--center_only",
action='store_true',
help="Switch for prediction from center crop alone instead of " +
"averaging predictions across crops (default).")
parser.add_argument(
"--images_dim",
default='256,256',
help="Canonical 'height,width' dimensions of input images.")
parser.add_argument(
"--mean_file",
default=os.path.join(pycaffe_dir,
'caffe/imagenet/ilsvrc_2012_mean.npy'),
help="Data set image mean of [Channels x Height x Width] dimensions " +
"(numpy array). Set to '' for no mean subtraction.")
parser.add_argument(
"--input_scale",
type=float,
help="Multiply input features by this scale to finish preprocessing.")
parser.add_argument(
"--raw_scale",
type=float,
default=255.0,
help="Multiply raw input by this scale before preprocessing.")
parser.add_argument(
"--channel_swap",
default='2,1,0',
help="Order to permute input channels. The default converts " +
"RGB -> BGR since BGR is the Caffe default by way of OpenCV.")
parser.add_argument(
"--ext",
default='jpg',
help="Image file extension to take as input when a directory " +
"is given as the input file.")
args = parser.parse_args()
image_dims = [int(s) for s in args.images_dim.split(',')]
mean, channel_swap = None, None
if args.mean_file:
mean = np.load(args.mean_file)
if args.channel_swap:
channel_swap = [int(s) for s in args.channel_swap.split(',')]
if args.gpu:
caffe.set_mode_gpu()
print("GPU mode")
else:
caffe.set_mode_cpu()
print("CPU mode")
# Make classifier.
classifier = caffe.Classifier(args.model_def,
args.pretrained_model,
image_dims=image_dims,
mean=mean,
input_scale=args.input_scale,
raw_scale=args.raw_scale,
channel_swap=channel_swap)
# Load numpy array (.npy), directory glob (*.jpg), or image file.
args.input_file = os.path.expanduser(args.input_file)
if args.input_file.endswith('npy'):
print("Loading file: %s" % args.input_file)
inputs = np.load(args.input_file)
elif os.path.isdir(args.input_file):
print("Loading folder: %s" % args.input_file)
inputs = [
caffe.io.load_image(im_f)
for im_f in glob.glob(args.input_file + '/*.' + args.ext)
]
else:
print("Loading file: %s" % args.input_file)
inputs = [caffe.io.load_image(args.input_file)]
print("Classifying %d inputs." % len(inputs))
# Classify.
start = time.time()
predictions = classifier.predict(inputs, not args.center_only)
print("Done in %.2f s." % (time.time() - start))
# Save
print("Saving results into %s" % args.output_file)
np.save(args.output_file, predictions)
# load the trained net
#
MODEL = '/u/amo-d0/grad/rbalten/research/deeptransient/src/caffemodels/deploy.prototxt'
PRETRAINED = '/u/amo-d0/grad/rbalten/research/deeptransient/src/caffemodels/transientneth.caffemodel'
MEAN = '../mean/transient_mean.binaryproto'
# load the mean image
blob = caffe.io.caffe_pb2.BlobProto()
file = open(MEAN, 'rb')
blob.ParseFromString(file.read())
means = caffe.io.blobproto_to_array(blob)
means = means[0]
caffe.set_mode_cpu()
net = caffe.Classifier(MODEL, PRETRAINED, caffe.TEST)
#
# process
#
ix = 0
db = lmdb.open(db_name)
# get all keys
with db.begin(write=False) as db_txn:
for (key, value) in db_txn.cursor():
im_datum = caffe.io.caffe_pb2.Datum()
im_datum.ParseFromString(value)
im = caffe.io.datum_to_array(im_datum)
# subtract mean & resize
f = StringIO()
PIL.Image.fromarray(a).save(f, fmt)
display(Image(data=f.getvalue()))
# Patching model to be able to compute gradients.
# Note that you can also manually add "force_backward: true" line to "deploy.prototxt".
model = caffe.io.caffe_pb2.NetParameter()
text_format.Merge(open(net_fn).read(), model)
model.force_backward = True
open('tmp.prototxt', 'w').write(str(model))
net = caffe.Classifier(
'tmp.prototxt',
param_fn,
mean=np.float32([104.0, 116.0,
122.0]), # ImageNet mean, training set dependent
channel_swap=(
2, 1,
0)) # the reference model has channels in BGR order instead of RGB
# a couple of utility functions for converting to and from Caffe's input image layout
def preprocess(net, img):
return np.float32(np.rollaxis(img, 2)[::-1]) - net.transformer.mean['data']
def deprocess(net, img):
return np.dstack((img + net.transformer.mean['data'])[::-1])
def objective_L2(dst):
print "Usage: python create_kaggle_submission_probability.py deploy.prototxt model.caffemodel mean.npy <test_data_dir>"
sys.exit()
DEPLOY_PROTOTXT = sys.argv[1] #deploy.prototxt
CAFFE_MODEL = sys.argv[2] #model.caffemodel
MEAN_FILE = sys.argv[3] #mean.npy
TEST_DATA_DIR = sys.argv[4]
MODEL_POSTFIX = CAFFE_MODEL.split(os.sep)[-2] # not safe way
BATCH_SIZE = 1 # edit it to fit your GPU memory
print "MODEL_POSTFIX", MODEL_POSTFIX
caffe.set_mode_gpu()
net = caffe.Classifier(DEPLOY_PROTOTXT,
CAFFE_MODEL,
mean=np.load(MEAN_FILE).mean(1).mean(1),
channel_swap=(2, 1, 0),
raw_scale=255,
image_dims=(256, 256))
def predict_proba(images_batch):
predictions = net.predict(images_batch)
#taking dog class predictions
batch_predictions = predictions[:, 1]
batch_predictions = batch_predictions[..., np.newaxis]
return batch_predictions #probability that the image is a dog (1 = dog, 0 = cat)
def main(argv):
pycaffe_dir = os.path.dirname(__file__)
parser = argparse.ArgumentParser()
# Required arguments: input and output files.
parser.add_argument("input_file", help="Input image, directory, or npy.")
parser.add_argument("output_file", help="Output npy filename.")
# Optional arguments.
parser.add_argument("--model_def",
default=os.path.join(pycaffe_dir,
"python/lenet.prototxt"),
help="Model definition file.")
parser.add_argument("--pretrained_model",
default=os.path.join(
pycaffe_dir, "python/lenet_iter_5000.caffemodel"),
help="Trained model weights file.")
parser.add_argument("--gpu",
action='store_true',
help="Switch for gpu computation.")
parser.add_argument(
"--center_only",
action='store_true',
help="Switch for prediction from center crop alone instead of " +
"averaging predictions across crops (default).")
parser.add_argument(
"--images_dim",
default='28,28',
help="Canonical 'height,width' dimensions of input images.")
parser.add_argument(
"--mean_file",
default=os.path.join(pycaffe_dir,
'caffe/imagenet/ilsvrc_2012_mean.npy'),
help="Data set image mean of H x W x K dimensions (numpy array). " +
"Set to '' for no mean subtraction.")
parser.add_argument(
"--input_scale",
type=float,
help="Multiply input features by this scale before input to net")
parser.add_argument(
"--channel_swap",
default='2,1,0',
help="Order to permute input channels. The default converts " +
"RGB -> BGR since BGR is the Caffe default by way of OpenCV.")
parser.add_argument(
"--ext",
default='jpg',
help="Image file extension to take as input when a directory " +
"is given as the input file.")
parser.add_argument("--labels_file",
default=os.path.join(pycaffe_dir, "mnist_words.txt"),
help="Readable label definition file.")
parser.add_argument(
"--print_results",
action='store_true',
help="Write output text to stdout rather than serializing to a file.")
parser.add_argument(
"--force_grayscale",
action='store_true',
help="Converts RGB images down to single-channel grayscale versions," +
"useful for single-channel networks like MNIST.")
args = parser.parse_args()
image_dims = [int(s) for s in args.images_dim.split(',')]
if args.force_grayscale:
channel_swap = None
mean = None
else:
channel_swap = [int(s) for s in args.channel_swap.split(',')]
mean = np.load(args.mean_file)
# Make classifier.
classifier = caffe.Classifier(
args.model_def,
args.pretrained_model,
image_dims=image_dims,
mean=mean,
input_scale=args.input_scale) #, channel_swap=channel_swap
if args.gpu:
print 'GPU mode'
print 'Get in while'
total_counter = 1
total_file_number = 100
sub_counter = 0
sub_file_number = 0
my_list_prob = []
my_list_label = []
file = open("prediction\\prediction_file_no_sub.txt", "r+")
while total_counter <= total_file_number:
my_list_prob = []
my_list_label = []
sub_counter = 0
while sub_counter <= sub_file_number:
print 'total_count: ' + '%d' % total_counter
# Load numpy array (.npy), directory glob (*.jpg), or image file.
FilePath = args.input_file + "test (%d)_%d.jpg" % (total_counter,
sub_counter)
print FilePath
FilePath = os.path.expanduser(FilePath)
if FilePath.endswith('npy'):
inputs = np.load(FilePath)
elif os.path.isdir(FilePath):
inputs = [
caffe.io.load_image(im_f)
for im_f in glob.glob(FilePath + '/*.' + args.ext)
]
else:
inputs = [caffe.io.load_image(FilePath)]
if args.force_grayscale:
inputs = [rgb2gray(input) for input in inputs]
#print inputs
inputs = np.asarray(inputs)
inputs = np.reshape(inputs, inputs.shape + (1, ))
print inputs.shape
print "Classifying %d inputs." % len(inputs)
print 'Get in Classify'
# Classify.
start = time.time()
scores = classifier.predict(inputs, not args.center_only).flatten()
print "Done in %.2f s." % (time.time() - start)
if args.print_results:
with open(args.labels_file) as f:
labels_df = pd.DataFrame([{
'synset_id':
l.strip().split(' ')[0],
'name':
' '.join(l.strip().split(' ')[1:]).split(',')[0]
} for l in f.readlines()])
labels = labels_df.sort('synset_id')['name'].values
indices = (-scores).argsort()[:5]
predictions = labels[indices]
meta = [(p, '%.5f' % scores[i])
for i, p in zip(indices, predictions)]
print meta
# Save
#np.save(args.output_file, predictions[0])
if float(meta[0][1]) > 0.4:
my_list_prob.append(meta[0][1])
my_list_label.append(meta[0][0])
#if (sub_counter!=0) and (sub_counter % sub_file_number) == 0:
if my_list_prob:
Max_num_index = int(my_list_prob.index(max(my_list_prob)))
else:
my_list_prob.append(meta[0][1])
my_list_label.append(meta[0][0])
Max_num_index = int(my_list_prob.index(max(my_list_prob)))
print my_list_label[Max_num_index]
save_str = "%d\t" % total_counter
file.write(save_str + str(my_list_label[Max_num_index]) + "\r\n")
sub_counter += 1
total_counter += 1
print total_counter
file.close()
# Setting this to CPU, but feel free to use GPU if you have CUDA installed
caffe.set_mode_cpu()
# Loading the Caffe model, setting preprocessing parameters
net = caffe.Net(model_prototxt, model_trained, caffe.TEST)
transformer = caffe.io.Transformer({'data': net.blobs['data'].data.shape})
transformer.set_mean('data', np.load(mean_path).mean(1).mean(1))
transformer.set_transpose('data', (2, 0, 1))
transformer.set_raw_scale('data', 255.0)
transformer.set_channel_swap('data', (2, 1, 0)) # RGB -> BGR
data_blob_shape = net.blobs['data'].data.shape
data_blob_shape = list(data_blob_shape)
#net.blobs['data'].reshape(batchsize, data_blob_shape[1], data_blob_shape[2], data_blob_shape[3])
"""
caffe.Classifier(model_prototxt, model_trained,
mean=np.load(mean_path).mean(1).mean(1),
channel_swap=(2,1,0),
raw_scale=255,
image_dims=(256, 256))
"""
def extract_feature(list_image_file):
start = datetime.now()
batchsize = len(list_image_file)
net.blobs['data'].reshape(batchsize, data_blob_shape[1],
data_blob_shape[2], data_blob_shape[3])
net.blobs['data'].data[...] = map(
lambda x: transformer.preprocess('data', caffe.io.load_image(x)),
import caffe
import argparse
import numpy as np
caffe.set_mode_gpu()
caffe.set_device(0)
parser = argparse.ArgumentParser()
parser.add_argument('source', nargs='+', help='list of image(s)')
parser.add_argument('model', help='model')
parser.add_argument('weights', help='pre-trained weights')
parser.add_argument('output', help='output')
args = parser.parse_args()
classifier = caffe.Classifier(args.model, args.weights)
inputs = [caffe.io.load_image(input) for input in args.source]
predictions = classifier.predict(inputs)
with open(args.output, 'w') as f:
for file, prediction in zip(args.source, predictions):
top_3_labels = np.argsort(prediction)[-1::-1][:3]
top_3_confidence = prediction[top_3_labels]
s = '{},{},{},{},{},{},{}\n'.format(
file, top_3_labels[0], top_3_confidence[0], top_3_labels[1],
top_3_confidence[1], top_3_labels[2], top_3_confidence[2])
f.write(s)