def show_progress(x, net, title=None, handle=False, dname='', CLC=True, IF_SAVE=False):
'''
Helper function to show intermediate results during the gradient descent.
:param x: vectorised image on which the gradient descent is performed
:param net: caffe.Classifier object defining the network
:param title: optional title of figuer
:param handle: obtional return of figure handle
:return: figure handle (optional)
'''
global index
disp_image = (x.reshape(*net.blobs['data'].data.shape)[0].transpose(1,2,0)[:,:,::-1]-x.min())/(x.max()-x.min())
if CLC:
clear_output()
if disp_image.shape[2] == 1:
plt.imshow(disp_image[:,:,0], cmap='gray')
if IF_SAVE:
pimg.imsave(dname+str(index)+'.png',disp_image[:,:,0], cmap='gray')
index += 1
else:
plt.imshow(disp_image)
if title != None:
ax = plt.gca()
ax.set_title(title)
f = plt.gcf()
display()
plt.show()
if handle:
return f
def match_score2prim(score,primlist,stafflist,barlist,myimg):
notelist = score.getElementsByClass('Part')[0].getElementsByClass('Measure')[0].getElementsByClass('Note')
#read all stem primitives
stemlist = []
for i in primlist:
if i.name == "stem":
stemlist.append(i)
stemlist.sort(key=lambda x:x.locbegin[0],reverse=False)
#for i in stemlist:
# print i.locbegin[0]
note_range = []
cols = range(barlist[1]-barlist[0])#10 pixels per column, overlap by 5
column_labels = [0 for i in range(len(cols))]
for i in range(len(notelist)):
dur_label = int(4*notelist[i].duration.quarterLength)
pitch_label = notelist[i].pitch
left,right = get_stem_box(stemlist[i])
#note_range.append([x,x2])
up = stafflist[0]-20
down = stafflist[4]+20
#print dur_label,left,right
cur = myimg[up:down,int(left):int(right)]
debugdir = "./debug/"
if not os.path.exists(debugdir):
os.makedirs(debugdir)
pngname = "./debug/"+str(dur_label)+"_"+str(int(left))+".png"
mpimg.imsave(pngname,cur)
for j in range(int(left)-barlist[0],int(right)-barlist[0]):
column_labels[j] = dur_label
reply = ""
for i in column_labels:
reply += str(i) + " "
reply += "\n"
return reply
def main():
# parse command line arguments
parser = argparse.ArgumentParser(description='Colorize pictures')
parser.add_argument('greyImage', help='png image to be coloured')
parser.add_argument('markedImage', help='png image with colour hints')
parser.add_argument('output', help='png output file')
parser.add_argument('-v', '--view', help='display image', action='store_true')
args = parser.parse_args()
# Note: when reading .png, division by 255. is not required
# Note: when reading .bmp, division by 255. is required
# TODO: make this more universal, i.e., support various image formats
# read images
greyImage = mpimg.imread(args.greyImage, format='png')
markedImage = mpimg.imread(args.markedImage, format='png')
# colorize
colouredImage = colorize(greyImage, markedImage)
# save output
mpimg.imsave(args.output,colouredImage, format='png')
# display output, if requested
if args.view:
plt.imshow(colouredImage)
plt.show()
def save_image(self, filename, mode='rgb', antialiased=False):
"""Save view from all panels to disk
Parameters
----------
filename: string
Path to new image file.
mode : string
Either 'rgb' (default) to render solid background, or 'rgba' to
include alpha channel for transparent background.
antialiased : bool
Antialias the image (see :func:`mayavi.mlab.screenshot`
for details; default False).
.. warning::
Antialiasing can interfere with ``rgba`` mode, leading to opaque
background.
Notes
-----
Due to limitations in TraitsUI, if multiple views or hemi='split'
is used, there is no guarantee painting of the windows will
complete before control is returned to the command line. Thus
we strongly recommend using only one figure window (which uses
a Mayavi figure to plot instead of TraitsUI) if you intend to
script plotting commands.
"""
im = self.screenshot(mode, antialiased)
imsave(filename, im)
def saveVideo(I, IDims, filename, FrameRate = 30, YCbCr = False, Normalize = False):
#Overwrite by default
if os.path.exists(filename):
os.remove(filename)
N = I.shape[0]
if YCbCr:
for i in range(N):
frame = np.reshape(I[i, :], IDims)
I[i, :] = ntsc2rgb(frame).flatten()
if Normalize:
I = I-np.min(I)
I = I/np.max(I)
for i in range(N):
frame = np.reshape(I[i, :], IDims)
frame[frame < 0] = 0
frame[frame > 1] = 1
mpimage.imsave("%s%i.png"%(TEMP_STR, i+1), frame)
if os.path.exists(filename):
os.remove(filename)
#Convert to video using avconv
command = [AVCONV_BIN,
'-r', "%i"%FrameRate,
'-i', TEMP_STR + '%d.png',
'-r', "%i"%FrameRate,
'-b', '30000k',
filename]
subprocess.call(command)
#Clean up
for i in range(N):
os.remove("%s%i.png"%(TEMP_STR, i+1))
def main():
# Check and parse arguments
if len(sys.argv) < 2:
print("Not enough arguments given. Need full path for the image file.")
sys.exit()
source = sys.argv[1]
im_colors = { 'r':0, 'g':1, 'b':2}
for filename in os.listdir(source):
filepath = source + filename
if os.path.splitext(filepath)[1] == '.jpg':
# Load image
im = mpim.imread(filepath)
im_color = im_colors[filename[-5]]
# Switch color channel
rows, cols = im.shape
im_colored = np.zeros((rows, cols, 3))
im_colored[:,:,im_color] = im / 255.
# Save colored image
mpim.imsave(os.path.splitext(filepath)[0] + '.png', im_colored,
format='PNG')
def _fetch_captchas_to_dir(directory, num=1):
plt.ion()
plt.show()
for i in range(num):
img = _captcha_provider.fetch()
plt.clf()
plt.axis('off')
plt.imshow(img)
# http://stackoverflow.com/questions/12670101/matplotlib-ion-function
# -fails-to-be-interactive
# https://github.com/matplotlib/matplotlib/issues/1646/
plt.show()
plt.pause(1e-2)
while True:
seq = input('[{}] Enter the char sequence: '.format(i + 1))
# To skip a CAPTCHA.
# Warning: skipping may reduce the quality of the training set.
if seq == _SEQ_SKIP:
break
seq = _captcha_provider.canonicalize_seq(seq)
if not _captcha_provider.is_valid_seq(seq):
print('Invalid sequence!')
else:
break
if seq == _SEQ_SKIP:
print('Skipped manually')
continue
path = os.path.join(directory, _add_suffix(seq))
if os.path.isfile(path):
print('Warning: char sequence already exists in dataset! Skipping')
else:
mpimg.imsave(path, img)
plt.ioff()
def display_image(image):
import io
from matplotlib.image import imsave
from IPython import display
buf = io.BytesIO()
imsave(buf, image)
return display.display_png(display.Image(buf.getvalue()))
def generate_label(num):
for i in range(num):
pagenum = i
print "Label page num",pagenum
infile = "./meas/"+"poly_44_"+str(pagenum)
#read svg file and assemble
svg_file = infile + ".svg"
primlist,stafflist,barlist= svg2primlist(svg_file)
primitive_assemble(primlist)
#read xml file
xml_file = infile + ".xml"
score = converter.parse(xml_file)
#score.show('text')
#cut out png, and write out label for each column
myimg = mpimg.imread(infile+'.png')
extraspace = 30
cur = myimg[stafflist[0]-extraspace:stafflist[4]+extraspace,barlist[0]:barlist[1]]
pngname = infile+"_s"+".png"
mpimg.imsave(pngname,cur)
column_labels,range_labels = match_score2prim(score,primlist,stafflist,barlist,myimg,extraspace)
labelfile = infile+"_col.txt"
f = open(labelfile,"w")
f.write(column_labels)
f.close()
rangefile = infile+"_s.txt"
f2 = open(rangefile,"w")
f2.write(range_labels)
f2.close()
def averager(imgpaths, width=500, height=600, alpha=False,
blur_edges=False, out_filename='result.png', plot=False):
size = (height, width)
images = []
point_set = []
for path in imgpaths:
img, points = load_image_points(path, size)
if img is not None:
images.append(img)
point_set.append(points)
ave_points = locator.average_points(point_set)
num_images = len(images)
result_images = np.zeros(images[0].shape, np.float32)
for i in xrange(num_images):
result_images += warper.warp_image(images[i], point_set[i],
ave_points, size, np.float32)
result_image = np.uint8(result_images / num_images)
mask = blender.mask_from_points(size, ave_points)
if blur_edges:
blur_radius = 10
mask = cv2.blur(mask, (blur_radius, blur_radius))
if alpha:
result_image = np.dstack((result_image, mask))
mpimg.imsave(out_filename, result_image)
if plot:
plt.axis('off')
plt.imshow(result_image)
plt.show()
def warp_report(in_file):
"""Plot the registration summary using nipy contours"""
mni_file = op.join(os.environ["FSL_DIR"],
"data/standard/MNI152_T1_1mm_brain.nii.gz")
mni_img = nib.load(mni_file)
mni_data, mni_aff = mni_img.get_data(), mni_img.get_affine()
sub_img = nib.load(in_file)
sub_data, sub_aff = sub_img.get_data(), sub_img.get_affine()
sub_data[sub_data < 1] = 0
kwargs = dict(draw_cross=False, annotate=False)
cut_coords = dict(x=(-45, -12, 12, 45),
y=(-55, -25, 5, 45),
z=(-30, -5, 20, 40))
colors = sns.color_palette("bright")
im_data = dict()
for axis in ["x", "y", "z"]:
f = plt.figure(figsize=(10, 2.5))
coords = cut_coords[axis]
slicer = viz.plot_anat(sub_data, sub_aff, slicer=axis,
cut_coords=coords, figure=f, **kwargs)
slicer.contour_map(mni_data, mni_aff, colors=colors)
fname = "slices_%s.png" % axis
f.savefig(fname, facecolor="k", edgecolor="k")
im_data[axis] = mplimg.imread(fname)
concat_data = [im_data[axis] for axis in ["x", "y", "z"]]
concat_data = np.concatenate(concat_data, axis=0)
mplimg.imsave("warp_report.png", concat_data)
return op.abspath("warp_report.png")
def imgHibrida():
raiz = os.getcwd()
filtro = gaussiana(9)
alinear(Image.open(raiz + "\human.png"), Image.open(raiz + "\cat.png"))
gato = mpimg.imread(raiz + "\cat.png")
print gato.shape
humano = mpimg.imread(raiz + "\humanAlign.png")
gatoConv = lowFilter(gato, filtro)
humanoConv = lowFilter(humano, filtro)
gatoHighConv = highFilter(gato, gatoConv)
plt.show()
plt.imshow(gatoHighConv)
plt.colorbar()
plt.title("Gat(Convolution with hp)")
plt.show()
plt.imshow(humanoConv)
plt.colorbar()
plt.title("Human(Convolution with lp)")
finalImage = gatoHighConv + humanoConv
normalizar(finalImage)
plt.show()
plt.imshow(finalImage)
plt.colorbar()
plt.title("Hybrid Image")
mpimg.imsave("HybridImage1.png", finalImage)
def post_process(self, sequence):
# postprocess a sequence in some way
sequence = np.array(sequence[self.params['n_steps']:])
save_generated_example('generated.wav',np.array(sequence))
mpimg.imsave('sequence.png',sequence)
def custom_imsave(path, image):
imsave(path, image, cmap=cm.Greys_r)
return
ret = image.copy()
ret = ret / (ret.max() / 255.0)
imsave(path, ret.astype(int))
def generate_tweet():
"""Generate a tweet and jpg.
Returns
-------
(status, jpg) : (str, str)
"""
imageid, fitsurl = select_random_image()
log.info('Opening {0}'.format(fitsurl))
fts = fits.open(fitsurl)
# Create the status message
imgtime = fts[0].header['IMG-TIME']
instrument = fts[0].header['INSTRUME'][8:]
exptime = fts[0].header['EXPTIME']
timestr = Time(imgtime).datetime.strftime('%d %b %Y, %H:%M UT').lstrip("0")
url = 'http://imagearchives.esac.esa.int/picture.php?/{0}'.format(imageid)
status = ('A new #Rosetta image was recently released!\n'
'⌚ {}.\n'
'📷 #{}.\n'
'⌛ {:.2f}s.\n'
'🔗 {}'.format(timestr, instrument, exptime, url))
# Create the cropped and scaled image
image_cropped = entropy_crop(fts[0].data, width=600, height=300)
image_scaled = scale_image(image_cropped, scale='linear',
min_percent=0.05, max_percent=99.95)
# Save the result as an image
image_fn = '/tmp/rosettabot.png'
log.info('Writing {0}'.format(image_fn))
imsave(image_fn, image_scaled, cmap=cm.gray)
return (status, image_fn)
def savePic2(fname, matrix, mode='normal'):
if mode == 'normal':
pimg.imsave(fname, matrix)
elif mode == 'gray':
pimg.imsave(fname, matrix, cmap=mlib.cm.gray)
else:
pass
def getImagePath(self, img_obj):
buffer = io.BytesIO()
mpimg.imsave(buffer,img_obj)
try:
mpimg.imsave(buffer,img_obj)
except:
print("Error: getImage method must return an image object")
return(buffer)
def _save_cm(output_cmap, cmap, format='png', n_colors=256):
""" Save the colormap of an image as an image file.
"""
# save the colormap
data = np.arange(0., n_colors) / (n_colors - 1.)
data = data.reshape([1, n_colors])
imsave(output_cmap, data, cmap=cmap, format=format)
def extract_hue(infile):
"""
Returns a hue greyscale image from an rgb image.
"""
rgb_image = imread(infile)
hsv_image = rgb_to_hsv(rgb_image)
hue_image = -hsv_image[:,:,0]
imsave(fname='%s-hue.png' %(infile.split('.')[0]),arr=hue_image,cmap='gray')
def render_offroad_tile(tile):
mask = vecttile_mask(tile)
heatmap = heatmap_array(tile_path(os.path.join(tiledir, 'strava'), tile, 'png'))
mod = adjust_alpha(heatmap.copy(), mask)
path = tile_path(os.path.join(tiledir, 'offroad'), tile, 'png')
prep_dirs(path)
image.imsave(path, mod)
return path
def preview(shape, solutions):
import os
remove("preview", "*.png")
for i, solution in enumerate(solutions):
image = matrix_to_image(shape, solution)
imsave(os.path.join("preview", "%d.png" % i), image)
if is_windows:
os.system("preview\\0.png")
def compare_image_lists(new_result, old_result, decimals):
fns = ['old.png', 'new.png']
num_images = len(old_result)
assert(num_images > 0)
for i in range(num_images):
mpimg.imsave(fns[0], np.loads(zlib.decompress(old_result[i])))
mpimg.imsave(fns[1], np.loads(zlib.decompress(new_result[i])))
assert compare_images(fns[0], fns[1], 10**(-decimals)) == None
for fn in fns: os.remove(fn)
def convert(image_file, outfile):
original_image = img.imread(image_file)
image = rgb_to_grayscale(original_image)
height, width = image.shape
print "Converting image ({0}x{1}) to graph...".format(width, height)
img.imsave("tmp/grayscale.png", image, cmap="gray")
graph = {} # graph has structure vertex_id -> edges
counter = 0
for i in xrange(height):
for j in xrange(width):
my_id = str(i * width + j)
edges = []
# create edge between adjaceny pixels
top_left = get_edge(i, j, i-1, j-1, height, width, image)
top = get_edge(i, j, i-1, j, height, width, image)
top_right = get_edge(i, j, i-1, j+1, height, width, image)
left = get_edge(i, j, i, j-1, height, width, image)
right = get_edge(i, j, i, j+1, height, width, image)
bottom_left = get_edge(i, j, i+1, j-1, height, width, image)
bottom = get_edge(i, j, i+1, j, height, width, image)
bottom_right = get_edge(i, j, i+1, j+1, height, width, image)
edges.append(top_left)
edges.append(top)
edges.append(top_right)
edges.append(left)
edges.append(right)
edges.append(bottom_left)
edges.append(bottom)
edges.append(bottom_right)
# remove out of range edges
lst = range(len(edges))
lst.reverse()
for k in lst:
if edges[k] == None:
edges.pop(k)
graph[my_id] = edges
counter +=1
# super source and super sink
source_edges = []
for i in xrange(height):
for j in xrange(width):
v_id = str(i * width + j)
source_edges.append([v_id, source_weight(i, j, image)])
graph[v_id].append(["t", sink_weight(i, j, image)])
graph["s"] = source_edges
graph["t"] = []
# write the outfile
outfile = open(outfile, "w")
for u, edges in graph.iteritems():
outfile.write(json.dumps(u) + "\t" + json.dumps(edges) + "\n")
def write_image(arr, name, suffix='.png', mode=None): from ..base.plots import colormap_for_mode if mode==None: cmap = colormap_for_mode(res['mode']) else: cmap = colormap_for_mode(mode) from ..base import make_filename imagefile = make_filename(args_file, name, suffix, work_dir) import matplotlib.image as pltimg pltimg.imsave(imagefile, arr, dpi=144, cmap=cmap)
def inner(self, *args, **kwargs):
if not self.do_plot or self.filepath is None:
pass
if len(args) > 1 and args[1] is 'save' and self.filepath is not None:
filename = self.filepath.format(self.counter - 1)
mpimg.imsave(filename, args[0])
print(filename)
if self.do_plot:
func(self, *args, **kwargs)
self.counter += 1
def post_process(self, sequence):
# postprocess a sequence in some way
sequence = np.array(sequence[self.params['n_steps']:])
mpimg.imsave('sequence.png', sequence, cmap=plt.cm.gray)
# ex. convert each vec to binary and print
fig = plt.figure(figsize=(10,10))
sub = fig.add_subplot(111)
sub.matshow(sequence, cmap=plt.cm.gray)
plt.show()
def save_w_as_image( X, in_w, in_h, out_w, out_h, outfile, mode, dpi=288, verbose=True, lms=False, ): assert len(X.shape) == 2, 'imagetiles: size error' if mode=='rgb': vis_ch = 3 elif mode=='rg' or mode=='rg_vs_b': vis_ch = 2 else: vis_ch = 1 x = out_h y = out_w _h = in_h _w = in_w xy, hw = X.shape assert (xy == x*y) and (hw == _h*_w*vis_ch), 'imagetiles: size error' if mode=='rgb': X = stack_matrix(X, _w, y) elif mode=='rg' or mode=='rg_vs_b': X = stack_matrix(X, _w, y, mode=mode) frame = 1 if mode=='lum': Y = N.ones((frame+x*(_h+frame), frame+y*(_w+frame))) * .05 else: Y = N.ones((frame+x*(_h+frame), frame+y*(_w+frame), 3)) * .05 image_id = 0 for xx in range(x): for yy in range(y): if image_id >= xy: break beginH, beginW = frame+xx*(_h+frame), frame+yy*(_w+frame) if mode=='rgb': tile = N.array([N.reshape(X[image_id].T[0], (_h, _w)), N.reshape(X[image_id].T[1], (_h, _w)), N.reshape(X[image_id].T[2], (_h, _w))]).T elif mode=='rg': tile = N.array([N.reshape(X[image_id].T[0], (_h, _w)), N.reshape(X[image_id].T[1], (_h, _w)), N.zeros((_h, _w))]).T elif mode=='rg_vs_b': tile = N.array([N.reshape(X[image_id].T[0], (_h, _w)), N.reshape(X[image_id].T[0], (_h, _w)), N.reshape(X[image_id].T[1], (_h, _w))]).T else: tile = N.reshape(X[image_id], (_h, _w)) Y[beginH : beginH+_h, beginW : beginW+_w] = tile image_id += 1 cmap = colormap_for_mode(mode) Y = normalize_color(Y) pltimg.imsave(outfile, Y, dpi=dpi, cmap=cmap) if (verbose): print 'file:', outfile, 'written.'
def write(self, imagearr, file_name=None):
if file_name != None:
self.file_name = file_name
elif self.file_name != None:
pass
else:
raise Exception(" %s , Undefined file name: " %
sys._getframe().f_code.co_name)
imagearr.un_normalize()
imsave(self.file_name, imagearr.get_uint_array(tdepth=8))
def slice_save(self, outputFile):
'''
Processes/saves a single slice.
ARGS
o outputFile
The output filename to save the slice to.
'''
if self._verbose:
self._log('Output file: %s\n' % outputFile)
image.imsave(outputFile, self._2Dslice, cmap = cm.Greys_r)
def save_masks_to_dir(dataset, all_masks):
os.mkdir("output/%s" % dataset.name)
for t in range(len(dataset.time)):
os.mkdir("output/%s/time%02d" % (dataset.name, t))
for s in range(len(dataset.slices)):
mask = all_masks[t][s]
image.imsave("output/%s/time%02d/slice%02d_mask.png" % (dataset.name, t, s), mask)
eroded = binary_erosion(mask)
hollow_mask = np.where(eroded, 0, mask)
colorimg = cv2.cvtColor(dataset.images[s][t], cv2.COLOR_GRAY2RGB)
colorimg = colorimg.astype(np.uint8)
colorimg[hollow_mask != 0] = [255, 0, 255]
image.imsave("output/%s/time%02d/slice%02d_color.png" % (dataset.name, t, s), colorimg)
def process_image(img):
global counter
mplimg.imsave(
os.path.join(output_dir, "output_{:05d}.jpg".format(counter)), img)
counter += 1
return img
keypoint_coord3d_v = sess.run(keypoint_coord3d_tf,
feed_dict={image_tf: image_v})
# Classifying based on Geometry
if args.solve_by == 0:
score_label = predict_by_geometry(keypoint_coord3d_v,
known_finger_poses,
args.threshold)
# Classifying based on Neural networks
elif args.solve_by == 1:
score_label = predict_by_neural_network(keypoint_coord3d_v,
known_finger_poses,
args.pb_file,
args.threshold)
# Classifying based on SVM
elif args.solve_by == 2:
score_label = predict_by_svm(keypoint_coord3d_v,
known_finger_poses, args.svc_file)
font = cv2.FONT_HERSHEY_SIMPLEX
cv2.putText(image_raw, score_label, (10, 200), font, 1.0, (255, 0, 0),
2, cv2.LINE_AA)
file_name = os.path.basename(img_name)
file_name_comp = file_name.split('.')
file_save_path = os.path.join(output_path,
"{}_out.png".format(file_name_comp[0]))
mpimg.imsave(file_save_path, image_raw)
print('{} --> {}\n\n'.format(file_name, score_label))
def execute_image_pipeline(self, visualize=False):
images = glob.glob(self.output_images_path + 'undist_*.jpg')
for idx, fname in enumerate(images):
image_fname = ntpath.basename(fname)
print("Processing", fname)
output_image = os.path.join(self.output_images_path, image_fname)
image = mpimg.imread(output_image)
self.image_shape = image.shape
thresholded_binary, warped, out_img, ploty, left_fitx, right_fitx = \
self.pipeline(
image,
sobel_kernel_size=self.sobel_kernel_size,
sx_thresh=self.sx_thresh,
sy_thresh=self.sy_thresh,
s_thresh=self.s_thresh,
mag_thresh=self.mag_thresh,
dir_thresh=self.dir_thresh,
test_image_pipeline=True,
visualize=True)
result, newwarp = self.project_back(image, thresholded_binary, warped, ploty, left_fitx, right_fitx)
result = self.plot_dashboard(result, ploty, left_fitx, right_fitx, newwarp)
output_image_thres = os.path.join(self.output_images_path, 'thres_' + image_fname)
mpimg.imsave(output_image_thres, thresholded_binary, cmap=cm.gray)
output_image_warp = os.path.join(self.output_images_path, 'warp_' + 'thres_' + image_fname)
mpimg.imsave(output_image_warp, warped, cmap=cm.gray)
output_image_out = os.path.join(self.output_images_path, 'out_' + 'warp_' + 'thres_' + image_fname)
mpimg.imsave(output_image_out, out_img)
output_image_result = os.path.join(self.output_images_path, 'result_' + 'out_' + 'warp_' + 'thres_' + image_fname)
mpimg.imsave(output_image_result, result)
if visualize:
# Plot the result
f, (ax1, ax2) = plt.subplots(1, 2, figsize=(24, 9))
f.tight_layout()
ax1.imshow(image)
ax1.set_title('Undistorted Image', fontsize=30)
ax2.imshow(thresholded_binary, cmap='gray')
ax2.set_title('Thresholded Result', fontsize=30)
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
ax2.imshow(warped, cmap='gray')
ax2.set_title('Warped Result', fontsize=30)
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
ax2.imshow(out_img, cmap='gray')
ax2.set_title('Line fit', fontsize=30)
ax2.plot(left_fitx, ploty, color='yellow')
ax2.plot(right_fitx, ploty, color='yellow')
plt.subplots_adjust(left=0., right=1, top=0.9, bottom=0.)
ax1.imshow(result)
ax1.set_title('Output Image', fontsize=30)
def find_lines_sliding_window(binary_warped, window_search, display):
histogram = np.sum(binary_warped[int(binary_warped.shape[0] / 2):, :],
axis=0)
out_img = np.dstack((binary_warped, binary_warped, binary_warped)) * 255
# we need max for each half of the histogram. the example above shows how
# things could be complicated if didn't split the image in half
# before taking the top 2 maxes
midpoint = np.int(histogram.shape[0] / 2)
leftx_base = np.argmax(histogram[:midpoint])
rightx_base = np.argmax(histogram[midpoint:]) + midpoint
# Choose the number of sliding windows
# this will throw an error in the height if it doesn't evenly divide the img height
nwindows = 9
# Set height of windows
window_height = np.int(binary_warped.shape[0] / nwindows)
# Identify the x and y positions of all nonzero pixels in the image
nonzero = binary_warped.nonzero()
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
# Current positions to be updated for each window
leftx_current = leftx_base
rightx_current = rightx_base
# Set the width of the windows +/- margin
margin = 100
# Set minimum number of pixels found to recenter window
minpix = 50
# Create empty lists to receive left and right lane pixel indices
left_lane_inds = []
right_lane_inds = []
# Step through the windows one by one
for window in range(nwindows):
# Identify window boundaries in x and y (and right and left)
win_y_low = int(binary_warped.shape[0] - (window + 1) * window_height)
win_y_high = int(binary_warped.shape[0] - window * window_height)
win_xleft_low = leftx_current - margin
win_xleft_high = leftx_current + margin
win_xright_low = rightx_current - margin
win_xright_high = rightx_current + margin
# Draw the windows on the visualization image
cv2.rectangle(out_img, (win_xleft_low, win_y_low),
(win_xleft_high, win_y_high), (0, 255, 0), 3)
cv2.rectangle(out_img, (win_xright_low, win_y_low),
(win_xright_high, win_y_high), (0, 255, 0), 3)
# Identify the nonzero pixels in x and y within the window
good_left_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) &
(nonzerox >= win_xleft_low) &
(nonzerox < win_xleft_high)).nonzero()[0]
good_right_inds = ((nonzeroy >= win_y_low) & (nonzeroy < win_y_high) &
(nonzerox >= win_xright_low) &
(nonzerox < win_xright_high)).nonzero()[0]
# Append these indices to the lists
left_lane_inds.append(good_left_inds)
right_lane_inds.append(good_right_inds)
# If you found > minpix pixels, recenter next window on their mean position
if len(good_left_inds) > minpix:
leftx_current = np.int(np.mean(nonzerox[good_left_inds]))
if len(good_right_inds) > minpix:
rightx_current = np.int(np.mean(nonzerox[good_right_inds]))
# Concatenate the arrays of indices
left_lane_inds = np.concatenate(left_lane_inds)
right_lane_inds = np.concatenate(right_lane_inds)
# Extract left and right line pixel positions
leftx = nonzerox[left_lane_inds]
lefty = nonzeroy[left_lane_inds]
rightx = nonzerox[right_lane_inds]
righty = nonzeroy[right_lane_inds]
# Fit a second order polynomial to each
left_fit = np.polyfit(lefty, leftx, 2)
right_fit = np.polyfit(righty, rightx, 2)
# Generate x and y values for plotting
ploty = np.linspace(0, binary_warped.shape[0] - 1, binary_warped.shape[0])
left_fitx = left_fit[0] * ploty**2 + left_fit[1] * ploty + left_fit[2]
right_fitx = right_fit[0] * ploty**2 + right_fit[1] * ploty + right_fit[2]
nonzero = binary_warped.nonzero()
nonzeroy = np.array(nonzero[0])
nonzerox = np.array(nonzero[1])
margin = 100
left_lane_inds = (
(nonzerox >
(left_fit[0] *
(nonzeroy**2) + left_fit[1] * nonzeroy + left_fit[2] - margin)) &
(nonzerox <
(left_fit[0] *
(nonzeroy**2) + left_fit[1] * nonzeroy + left_fit[2] + margin)))
right_lane_inds = (
(nonzerox >
(right_fit[0] *
(nonzeroy**2) + right_fit[1] * nonzeroy + right_fit[2] - margin)) &
(nonzerox <
(right_fit[0] *
(nonzeroy**2) + right_fit[1] * nonzeroy + right_fit[2] + margin)))
# Again, extract left and right line pixel positions
leftx = nonzerox[left_lane_inds]
lefty = nonzeroy[left_lane_inds]
rightx = nonzerox[right_lane_inds]
righty = nonzeroy[right_lane_inds]
# Fit a second order polynomial to each
left_fit = np.polyfit(lefty, leftx, 2)
right_fit = np.polyfit(righty, rightx, 2)
if display:
out_img[lefty, leftx] = [255, 0, 0]
out_img[righty, rightx] = [0, 0, 255]
# Plots the left and right polynomials on the lane lines
plt.plot(left_fitx, ploty, color='yellow')
plt.plot(right_fitx, ploty, color='yellow')
plt.imshow(out_img)
mpimg.imsave('./output_images/10-window_search.jpg', out_img)
return left_fit, right_fit, leftx, lefty, rightx, righty, window_search
figure.append(temp)
print("figure:", figure)
# print(img.shape)
counter = img[position[1]:position[1] + position[3] + 1,
position[0]:position[0] + position[2] + 1]
counter = rgb2gray(counter)
#plt.imshow(counter, cmap='gray', vmin=0, vmax=255)
# plt.show()
for i in range(5):
img_figure = img[figure[i][1]:figure[i][1] + figure[i][3] + 1,
figure[i][0]:figure[i][0] + figure[i][2] + 1]
img_figure = rgb2gray(img_figure)
mpimg.imsave(output_path + str(figure_reading[i]) + '/' +
str(index) + '.jpg',
img_figure,
cmap='gray')
index += 1
#plt.imshow(img_figure, cmap='gray', vmin=0, vmax=255)
# plt.show()
"""
subprocess.call("rm -rf " + data_folder + output_all, shell=True)
f = open(data_folder + output_all,'w')
for path,dir_list,file_list in g:
for file_name in file_list:
f.write(os.path.join(path, file_name))
f.write('\n')
image_path = "../Linux/data/11_3.jpg"
img = rgb2gray(mpimg.imread(image_path))
def uploader():
def load(fname):
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
import matplotlib.image as img
import math as m
import copy
dic={}
M = img.imread(fname)
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+fname, M)
dic['M']=M
W, H = M.shape[:2]
dic['W']=W
dic['H']=H
NEW = np.random.rand(W,H)
dic['NEW']=NEW
IMAG= np.random.rand(W,H)
dic['IMAG']=IMAG
FB=(1.0/273)*np.array([ [1,4,7,4,1],
[4,16,26,16,4],
[7,26,41,26,7],
[4,16,26,16,4],
[1,4,7,4,1]
])
FS=(1.0)*np.array([
[0,-1,0],
[-1,5,-1],
[0,-1,0],
])
dic['FB']=FB
dic['FS']=FS
return dic
def convertGray(fname):
import copy
import numpy as np
d=load(fname)
new=copy.deepcopy(d['NEW'])
w=copy.deepcopy(d['W'])
h=copy.deepcopy(d['H'])
m=copy.deepcopy(d['M'])
for i in range(w):
for j in range(h):
avg =float((m[i][j][0]*0.299)+(m[i][j][1]*0.587)+(m[i][j][2]*0.114))
#avg =(m[i][j][0]+m[i][j][1]+m[i][j][2])/3
new[i][j] = avg
return copy.deepcopy(new)
def blur(fname):
import numpy as np
import copy
d=load(fname)
m=copy.deepcopy(d['M'])
Xw=Xw=m.shape[1]
Xh=m.shape[0]
newss=np.zeros((Xh,Xw,3))
F=copy.deepcopy(d['FB'])
g=sum(sum(F))
Fh=F.shape[0]
Fw=F.shape[1]
H=(Fh-1)//2
W=(Fw-1)//2
for i in range(H,Xh-H):
for j in range(W,Xw-W):
s1=0
s2=0
s3=0
for k in range(-H,H+1):
for l in range(-W,W+1):
a1=m[i+k,j+l,0]
a2=m[i+k,j+l,1]
a3=m[i+k,j+l,2]
p1=F[H+k,W+l]
p2=F[H+k,W+l]
p3=F[H+k,W+l]
s1+=(p1*a1)
s2+=(p2*a2)
s3+=(p3*a3)
newss[i,j,0]=s1
newss[i,j,1]=s2
newss[i,j,2]=s3
return copy.deepcopy(newss/255)
def sharpen(fname):
import numpy as np
import copy
d=load(fname)
m=copy.deepcopy(d['M'])
Xw=Xw=m.shape[1]
Xh=m.shape[0]
newsss=np.zeros((Xh,Xw,3))
F=copy.deepcopy(d['FS'])
g=sum(sum(F))
Fh=F.shape[0]
Fw=F.shape[1]
H=(Fh-1)//2
W=(Fw-1)//2
for i in range(H,Xh-H):
for j in range(W,Xw-W):
s1=0
s2=0
s3=0
for k in range(-H,H+1):
for l in range(-W,W+1):
a1=m[i+k,j+l,0]
a2=m[i+k,j+l,1]
a3=m[i+k,j+l,2]
p1=F[H+k,W+l]
p2=F[H+k,W+l]
p3=F[H+k,W+l]
s1+=(p1*a1)
s2+=(p2*a2)
s3+=(p3*a3)
newsss[i,j,0]=s1
newsss[i,j,1]=s2
newsss[i,j,2]=s3
return copy.deepcopy(newsss)
def histogramequalisation(fname):
import copy
import numpy as np
new=convertGray(fname)
pmf={}
cmf={}
d=load(fname)
w=copy.deepcopy(d['W'])
h=copy.deepcopy(d['H'])
for i in range(w):
for j in range(h):
if new[i][j] in pmf:
pmf[new[i][j]]+=1
else:
pmf[new[i][j]]=1
for i in pmf:
pmf[i]=pmf[i]/(w*h)
s=0
for i in pmf:
if s==0:
s+=pmf[i]
cmf[i]=pmf[i]
else:
s+=pmf[i]
cmf[i]=s
for i in cmf:
cmf[i]=int(cmf[i]*255)
newpmf={}
for i in pmf:
newpmf[cmf[i]]=pmf[i]*(w*h)
pix={}
for i in pmf:
pix[i]=cmf[i]
chan = np.random.rand(w,h)
for i in range(w):
for j in range(h):
chan[i][j]=pix[new[i][j]]
return copy.deepcopy(chan)
def negative(fname):
import numpy as np
import copy
from copy import deepcopy
d=load(fname)
w=copy.deepcopy(d['W'])
h=copy.deepcopy(d['H'])
neg=np.zeros((w,h))
new=convertGray(fname)
for i in range(w):
for j in range(h):
neg[i][j]=255-new[i][j]
return deepcopy(neg)
def logtransform(fname):
import numpy as np
from copy import deepcopy
import math as m
import copy
d=load(fname)
w=copy.deepcopy(d['W'])
h=copy.deepcopy(d['H'])
log=np.zeros((w,h))
new=convertGray(fname)
for i in range(w):
for j in range(h):
log[i][j]=0.004*m.log(new[i][j]+1)
return deepcopy(log)
def gamma(fname):
import numpy as np
import copy
from copy import deepcopy
d=load(fname)
w=copy.deepcopy(d['W'])
h=copy.deepcopy(d['H'])
gama=np.zeros((w,h))
new=convertGray(fname)
for i in range(w):
for j in range(h):
gama[i][j]=new[i][j]**1.5
return deepcopy(gama)
def linear(fname):
import numpy as np
import copy
from copy import deepcopy
d=load(fname)
w=copy.deepcopy(d['W'])
h=copy.deepcopy(d['H'])
lin=np.zeros((w,h))
new=convertGray(fname)
for i in range(w):
for j in range(h):
lin[i][j]=new[i][j]*25+1
return deepcopy(lin)
def MedianFilter(image, filter_size):
import numpy
result = numpy.zeros((len(image),len(image[0])))
filter_range = filter_size // 2
garbage = []
for i in range(len(image)):
for j in range(len(image[0])):
for z in range(filter_size):
if i + z - filter_range < 0 or i + z +filter_range > len(image) - 1:
for c in range(filter_size):
garbage.append(0)
else:
if j + z - filter_range < 0 or j + filter_range > len(image[0]) - 1:
garbage.append(0)
else:
for k in range(filter_size):
garbage.append(image[i + z - filter_range][j + k - filter_range])
garbage.sort()
result[i][j] = garbage[len(garbage) // 2]
garbage = []
return result
def Mean(image, filter_size):
import numpy
result = numpy.zeros((len(image),len(image[0])))
filter_range = filter_size // 2
garbage = []
for i in range(len(image)):
for j in range(len(image[0])):
for z in range(filter_size):
if i + z - filter_range < 0 or i + z +filter_range > len(image) - 1:
for c in range(filter_size):
garbage.append(0)
else:
if j + z - filter_range < 0 or j + filter_range > len(image[0]) - 1:
garbage.append(0)
else:
for k in range(filter_size):
garbage.append(image[i + z - filter_range][j + k - filter_range])
m=sum(garbage)//(len(garbage))
result[i][j] = m
garbage = []
return result
def rgb2gray(rgb):
return np.dot(rgb[...,:3], [0.2989, 0.5870, 0.1140])
def BandB(new):
s=0
w, h = new.shape[:2]
for i in range(w):
for j in range(h):
s+=new[i][j]
threshold=s//(new.shape[0]*new.shape[1])
for i in range(w):
for j in range(h):
if new[i][j]>=threshold:
new[i][j]=1
else:
new[i][j]=0
return new
if request.method == 'POST':
import copy
import numpy as np
from PIL import Image
import matplotlib.pyplot as plt
import matplotlib.image as img
import math as m
import copy
import cv2 as cv
import io
f = request.files['file']
f.save(secure_filename(f.filename))
if request.form['submit_button'] == 'Gray':
new=convertGray(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'Gray'+str(f.filename), new,cmap='gray')
full_filename='Gray'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Blur':
new=blur(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'Blur'+str(f.filename), new)
full_filename='Blur'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Sharpen Kernel':
new=sharpen(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'S'+str(f.filename), new)
full_filename='S'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Histogram Equalisation':
new=histogramequalisation(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'Histo'+str(f.filename), new,cmap='gray')
full_filename='Histo'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Linear Transformation':
new=linear(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'linear'+str(f.filename), new,cmap='gray')
full_filename='linear'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Gamma Transformation':
new=gamma(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'gamma'+str(f.filename), new,cmap='gray')
full_filename='gamma'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Negative Transformation':
new=negative(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'negative'+str(f.filename), new,cmap='gray')
full_filename='negative'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Log Transformation':
new=logtransform(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'log'+str(f.filename), new,cmap='gray')
full_filename='log'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Median Filtering':
import numpy
imgi = img.imread(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+str(f.filename), imgi)
gray = rgb2gray(imgi)
arr = numpy.array(gray)
imgi = MedianFilter(arr, 3)
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'Med'+str(f.filename), imgi,cmap='gray')
full_filename='Med'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Mean Filtering':
import numpy
imgi = img.imread(str(f.filename))
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+str(f.filename), imgi)
gray = rgb2gray(imgi)
arr = numpy.array(gray)
imgi = Mean(arr, 3)
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'Mean'+str(f.filename), imgi,cmap='gray')
full_filename='Mean'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
elif request.form['submit_button'] == 'Black & White':
new=convertGray(str(f.filename))
new=BandB(new)
img.imsave('C:/Users/aditya/Desktop/cs300/static/'+'B&W'+str(f.filename), new,cmap='gray')
full_filename='B&W'+str(f.filename)
return render_template('image.html',name=full_filename,input=str(f.filename))
return 'file uploaded successfully'
sxbinary = np.zeros_like(scaled_sobel)
sxbinary[(scaled_sobel >= thresh_min) & (scaled_sobel <= thresh_max)] = 1
# Threshold color channel
s_thresh_min = 170
s_thresh_max = 255
s_binary = np.zeros_like(S)
s_binary[(S >= s_thresh_min) & (S <= s_thresh_max)] = 1
combined_binary = np.zeros_like(sxbinary)
combined_binary[(s_binary == 1) | (sxbinary == 1)] = 1
plt.figure(c)
plt.imshow(combined_binary, cmap='gray')
mpimg.imsave("output_images/" + fname + "_combBin.jpg",
combined_binary,
cmap='gray')
#trap = np.array([[[240,686],[1055,675],[690,450],[587,450]]], np.int32)
#wwww = np.array([[[100,img.shape[1]],[1100,img.shape[1]],[1100,0],[100,0]]], np.int32)
trap = np.array([[[568, 468], [715, 468], [1040, 680], [270, 680]]],
np.int32)
wwww = np.array([[[200, 0], [1000, 0], [1000, 680], [200, 680]]], np.int32)
src = np.float32(trap)
dst = np.float32(wwww)
M = cv2.getPerspectiveTransform(src, dst)
warped = cv2.warpPerspective(
combined_binary,
M, (combined_binary.shape[1], combined_binary.shape[0]),
flags=cv2.INTER_LINEAR)
for epoch in range(total_epochs):
for batch in range(total_batchs):
batch_x = train_x[batch * batch_size:(batch + 1) * batch_size]
noise = random_noise(batch_size)
sess.run(g_train, feed_dict={Z: noise})
sess.run(d_train, feed_dict={X: batch_x, Z: noise})
gl, dl = sess.run([g_loss, d_loss],
feed_dict={
X: batch_x,
Z: noise
})
if epoch % 2 == 0:
print("======= Epoch : ", epoch, " =======")
print("generator: ", -gl)
print("discriminator: ", -dl)
samples = 20
if epoch % 2 == 0:
sample_noise = random_noise(samples)
gen = sess.run(fake_x, feed_dict={Z: sample_noise})
for i in range(samples):
img = gen[i].reshape([256, 256, 3])
mpimg.imsave('./epoch/epoch' + str(epoch) + '_' + str(i) +
'.png',
img,
format='png')
% Remove the low frequencies from image2. The easiest way to do this is to
% subtract a blurred version of image2 from the original version of image2.
% This will give you an image centered at zero with negative values.
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
"""
lf2 = my_imfilter(image2, filter)
lf2 = my_imfilter(lf2, filter_transpose)
high_frequencies = image2 - lf2
"""
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
% Combine the high frequencies and low frequencies
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
"""
hybrid_image = low_frequencies + high_frequencies
#%% Visualize and save outputs
plt.figure(1)
plt.imshow(low_frequencies)
plt.figure(2)
plt.imshow(high_frequencies + 0.5)
vis = vis_hybrid_image(hybrid_image) #see function script vis_hybrid_image.py
plt.figure(3)
plt.imshow(vis)
mpimg.imsave('Results/low_frequencies.jpg', np.clip(low_frequencies, 0, 1))
mpimg.imsave('Results/high_frequencies.jpg',
np.clip(high_frequencies + 0.5, 0, 1.0))
mpimg.imsave('Results/em_hybrid_image.jpg', np.clip(hybrid_image, 0, 1))
mpimg.imsave('Results/em_hybrid_image_scales.jpg', np.clip(vis, 0, 1))
cv2.fillPoly(mask, vertices, ignore_mask_color)
masked_edges = cv2.bitwise_and(edges, mask)
# Define the Hough transform parameters
# Make a blank the same size as our image to draw on
rho = 2 # distance resolution in pixels of the Hough grid
theta = np.pi / 180 # angular resolution in radians of the Hough grid
threshold = 15 # minimum number of votes (intersections in Hough grid cell)
min_line_length = 40 #minimum number of pixels making up a line
max_line_gap = 20 # maximum gap in pixels between connectable line segments
line_image = np.copy(image) * 0 # creating a blank to draw lines on
# Run Hough on edge detected image
# Output "lines" is an array containing endpoints of detected line segments
lines = cv2.HoughLinesP(masked_edges, rho, theta, threshold, np.array([]),
min_line_length, max_line_gap)
# Iterate over the output "lines" and draw lines on a blank image
for line in lines:
for x1, y1, x2, y2 in line:
cv2.line(line_image, (x1, y1), (x2, y2), (255, 0, 0), 10)
# Create a "color" binary image to combine with line image
color_edges = np.dstack((edges, edges, edges))
# Draw the lines on the edge image
lines_edges = cv2.addWeighted(color_edges, 0.8, line_image, 1, 0)
mpimg.imsave('output.jpg', line_image)
plt.imshow(lines_edges)
plt.show()
import numpy as np
import matplotlib.image as img
import statistics as stits
a = img.imread("Sample.png")
h, w = a.shape[:2]
k = np.zeros([h, w, 3])
for i in range(0, h):
for j in range(0, w):
n = [a[i][j][0], a[i][j][1], a[i][j][2]]
m = int(stits.mean(n))
k[i][j][0] = int(m)
k[i][j][1] = int(m)
k[i][j][2] = int(m)
img.imsave('test.png', k)
signal = np.clip(signal, -500, 500)
# Calculate activation signal
return expit(signal)
def ReLU_function(signal):
# Return the activation signal
return np.maximum(0, signal)
################################################################################
# Load Input Image
################################################################################
img = Image.open(os.path.join(PATH_TO_TEST_IMAGES_DIR, PATH_TO_TEST_IMAGE))
# Save original image to the visualization folder
image.imsave(os.path.join(PATH_TO_INPUT_IMAGE, '{}.png'.format(0)), img)
input_image = np.array(img.getdata(), dtype=np.uint8)
input_image = np.resize(input_image, (img.size[1], img.size[0], 4))
input_image = input_image[:, :, 1]
################################################################################
# Convolution 1
################################################################################
conv1_output = np.empty((20, 20, 256))
for i in range(conv1_weights.shape[3]):
# Get the 9x9 kernel
extracted_filter = conv1_weights[:, :, :, i]
extracted_filter = np.squeeze(extracted_filter)
# Save image of the kernel
image.imsave(os.path.join(PATH_TO_CONV1_KERNEL, '{}.png'.format(i)),
if __name__ == "__main__":
labels = defaultdict(list)
with open('training_data/object-detection-crowdai/labels.csv',
'r') as csvfile:
label_data = csv.DictReader(csvfile)
for row in label_data:
if row['Label'] == 'Car':
x0, x1, y0, y1 = int(row['xmin']), int(row['ymin']), int(
row['xmax']), int(row['ymax'])
value = (min(x0, x1), max(x0, x1), min(y0, y1), max(y0, y1))
labels[row['Frame']].append(value)
filenames = glob.glob('training_data/object-detection-crowdai/*.jpg')
# out_path = 'training_data/udacity/cars/' #to void triggering by mistake and blowing away folder!!
if os.path.exists(out_path):
shutil.rmtree(out_path)
os.makedirs(out_path)
i = 0
for fname in filenames:
frame_name = os.path.basename(fname)
frame_labels = labels.get(frame_name)
if frame_labels is not None:
image = mpimg.imread(fname)
for label in frame_labels:
sub_image = image[label[2]:label[3], label[0]:label[1], :]
if sub_image.any():
outname = "{}/{}.png".format(out_path, i)
mpimg.imsave(outname, sub_image)
i += 1
#gray scale
import matplotlib.image as img
a=img.imread("sample.jpg")
w,h=a.shape[:2]
import numpy as np
b=np.zeros([w,h,3])
import statistics as stits
for i in range(0,w):
for j in range(0,h):
n=[a[i][j][0],a[i][j][1],a[i][j][2]]
k=int(stits.mean(n))
b[i][j][0]=int(k)
b[i][j][1]=int(k)
b[i][j][2]=int(k)
img.imsave('shine1.jpg',b)
img = mpimg.imread(img_path)
img = np.array(img)
mask = mpimg.imread(mask_path)
mask = np.array(mask)
#converting to grayscale
mask_gray = cv2.cvtColor(mask, cv2.COLOR_RGB2GRAY)
img_gray = cv2.cvtColor(img, cv2.COLOR_RGB2GRAY)
block_sizes = [5, 10, 15, 20, 25, 30]
SD_THs = [None, 1, 2, 3, 4]
pms = []
counter = 1
for size in block_sizes:
for SD_TH in SD_THs:
print("size :", size, " SD_TH:", SD_TH)
pm = CPF(img_gray, mask_gray, size, mean_threshold, SD_threshold,
min_pixel_distance, check_offset, SD_TH)
output = str(img_no) + "_" + str(counter) + ".png"
mpimg.imsave(output, pm, cmap="gray")
counter += 1
plt.show()
# -*- coding: utf-8 -*-
"""
Generates templates with dots for each domino
"""
import numpy as np
import matplotlib as mpl
from matplotlib.image import imsave
import cv2 as cv
import os
folder = "domino_templ"
base_images = []
final_images = []
for i in range(0, 7):
img = cv.imread(os.path.join(folder, "base/%i.png" % i))
if img is not None:
img = img[:, :, 1]
base_images.append(img)
#imsave(folder + '/0%i.png' %i, img)
for i in range(7):
for j in range(i + 1):
upper_half = base_images[i][:, :]
lower_half = base_images[j][::-1, ::-1]
img = np.vstack((upper_half[:100, :], lower_half[100:, :]))
imsave(folder + '/%i%i.png' % (j, i), 255 - img, cmap='binary')
def simulateByteErrors(I, IDims, fracbyte, doPlot=False):
"""
Create a version of the video which would occur from bit errors
:param I: NFrames x NPixels video array
:param IDims: Tuple of the dimensions of the pixels
:param fracbyte: The fraction of the bytes that are randomly changed
"""
TEMPAVI = "temp.avi"
LFAC = 400 #How many bytes to corrupt in a row
FLIPPROB = 0.5
FrameRate = 30
N = I.shape[0]
#Step 1: Output video as temporary files and convert to AVI
for i in range(N):
frame = np.reshape(I[i, :], IDims)
mpimage.imsave("%s%i.png" % (TEMP_STR, i + 1), frame)
#Convert to video using avconv
if os.path.exists(TEMPAVI):
os.remove(TEMPAVI)
command = [
AVCONV_BIN, '-r',
"%i" % FrameRate, '-i', TEMP_STR + '%d.png', '-r',
"%i" % FrameRate, '-b', '30000k', TEMPAVI
]
subprocess.call(command)
#Step 2: Read in AVI file as a binary file and corrupt it
fin = open(TEMPAVI, "rb")
b = fin.read()
fin.close()
#Cleanup leftover files
for i in range(N):
os.remove("%s%i.png" % (TEMP_STR, i + 1))
os.remove(TEMPAVI)
b = bytearray(b)
i = 15000 #Make sure the header is skipped
lam = LFAC / fracbyte
ilocs = []
while i < len(b):
i += int(np.random.exponential(lam))
ilocs.append(i)
NumFlipped = 0
for i in ilocs:
for k in range(i, min(i + LFAC, len(b))):
if np.random.rand() < FLIPPROB:
b[k] = np.random.randint(255)
NumFlipped += 1
if doPlot:
#For debugging
plt.stem(np.array(ilocs), np.ones(len(ilocs)))
plt.title("fracbyte = %g, %g flipped" %
(fracbyte, float(NumFlipped) / len(b)))
plt.show()
fout = open(TEMPAVI, "wb")
fout.write(b)
fout.close()
#Step 3: Use avconv to extract the corrupted frames
command = [AVCONV_BIN, '-i', TEMPAVI, '-f', 'image2', TEMP_STR + '%d.png']
subprocess.call(command)
os.remove(TEMPAVI)
i = 1
frames = []
IDimsRet = IDims
while os.path.exists("%s%d.png" % (TEMP_STR, i)):
filename = "%s%d.png" % (TEMP_STR, i)
F = scipy.misc.imread(filename)
IDimsRet = F.shape
#os.remove(filename)
frames.append(F.flatten())
i += 1
ret = np.array(frames) / 255.0
return (ret, IDimsRet)
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
@author: Wenxuan Liu 805152602
"""
import matplotlib.pyplot as plt
import matplotlib.image as mpimg
import numpy as np
if __name__ == "__main__":
img1 = mpimg.imread('g.jpg').copy()
img2 = mpimg.imread('h.jpg').copy()
img1 = img1 / 255.0
img2 = img2 / 255.0
a, b, c = img1.shape
img3 = np.empty([a, b, c])
for i in range(a):
for j in range(b):
for k in range(c):
img3[i, j, k] = abs(img2[i, j, k]-img1[i, j, k]) #find the difference
plt.axis('off')
plt.imshow(img3)
mpimg.imsave('i.jpg', img3)
plt.show()
for idx, fname in enumerate(images):
test_img = mpimg.imread(fname)
# Find_cars on test images
box_list, find_cars_img = find_cars(test_img,
ystart=ystart,
ystop=ystop,
color_space=color_space,
scale=scale,
svc=svc,
X_scaler=X_scaler,
orient=orient,
pix_per_cell=pix_per_cell,
cell_per_block=cell_per_block,
spatial_size=spatial_size,
hist_bins=hist_bins)
mpimg.imsave('./output_images/find_cars_' + str(idx) + '.jpg',
find_cars_img)
# Add heat to each box in box list
heat = np.zeros_like(test_img[:, :, 0]).astype(np.float)
heat = add_heat(heat, box_list)
# Visualize the heatmap when saving
heatmap = np.clip(heat, 0, 255)
mpimg.imsave('./output_images/add_heat_' + str(idx) + '.jpg',
heatmap,
cmap='hot')
# Apply threshold to help remove false positives
heat = apply_threshold(heat, threshold=threshold)
# Visualize the heatmap when saving
heatmap = np.clip(heat, 0, 255)
mpimg.imsave('./output_images/apply_threshold_' + str(idx) + '.jpg',
heatmap,
cmap='hot')
"""
The image "img" will be repeated n times in
vertical and m times in horizontal direction.
"""
if n == 1:
tiled_img = img
else:
lst_imgs = []
for i in range(n):
lst_imgs.append(img)
tiled_img = np.concatenate(lst_imgs, axis=1)
if m > 1:
lst_imgs = []
for i in range(m):
lst_imgs.append(tiled_img)
tiled_img = np.concatenate(lst_imgs, axis=0)
return tiled_img
basic_pattern = mpimg.imread('imag_tile_explanation.png')
decorators_img = imag_tile(basic_pattern, 3, 3)
print decorators_img
plt.axis("off")
plt.imshow(decorators_img)
cropped = basic_pattern[90:150, 50:120]
# print(at_img.shape)
mpimg.imsave('/home/halk/Pictures/decorators_with_at.png', cropped)
mpimg.imsave('/home/halk/Pictures/decorators_with_at1.png', decorators_img)
if i < 2:
continue
fname = x.split(' ')[0]
landmarks = x.split(' ')[1:]
landmarks = [int(a) for a in landmarks if len(a) > 0]
if len(landmarks) == 10:
nose_to_left_eye = landmarks[0] - landmarks[4]
eye_distance = landmarks[0] - landmarks[2]
rotation = nose_to_left_eye / float(eye_distance)
# print('%s %.04f' % (fname, rotation))
if (rotation > 0.35) and (rotation < 0.65):
eye_height = (landmarks[1] + landmarks[3]) // 2
mouth_height = (landmarks[7] + landmarks[9]) // 2
nose_height = landmarks[5]
# origin is top left
thirds_height = mouth_height - eye_height
top = eye_height - thirds_height
bottom = nose_height + thirds_height
left = landmarks[4] - thirds_height
right = landmarks[4] + thirds_height
# nose_to_right_eye = landmarks[2] - landmarks[4]
img = imread('%s/%s' % (src_folder, fname))
crop = img[top:bottom, left:right]
imsave('%s/%s' % (dst_folder, fname), crop)
# img.close()
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Sun Sep 15 15:07:42 2019
@author: nguyent
"""
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.image as mpg
largeur = 600
longueur = int(largeur * 2 / 3)
c = int(largeur * 1 / 3)
im = np.zeros((longueur, largeur, 3), dtype=np.uint8)
im[:, 0:c:1] = [5, 20, 64]
im[:, c:2 * c:1] = [255, 255, 255]
im[:, 2 * c:3 * c:1] = [236, 25, 32]
plt.imshow(im)
mpg.imsave('Flag/DrapeauFrancais.png', im)
x0 = start.real - (radius / 2) + (j * radius / nb_pxls)
y0 = start.imag - (radius / 2) + (i * radius / nb_pxls)
z0 = complex(x0, y0)
# gray = (MAX - how_long_within(z0, MAX))/MAX
# color = (gray, gray, gray)
index = MAX - how_long_within(z0, MAX)
color = COLORS[index]
mtrx[i, j] = color
return mtrx
# mtrx[i,j] =
if __name__ == "__main__":
parser = argparse.ArgumentParser(description='Draw mandelbrot set')
parser.add_argument('nb_pixels',
type=int,
help='size in pixels of the square image')
parser.add_argument('xc', type=float, help='real part of the start')
parser.add_argument('yc', type=float, help='imaginary part of the start')
parser.add_argument('radius',
type=float,
help='enter smaller numbers to zoom near')
args = parser.parse_args()
nb_pixels = args.nb_pixels
xc = args.xc
yc = args.yc
radius = args.radius
res = mandelbrot_image_matrix(nb_pixels, complex(xc, yc), radius)
mpimg.imsave('tes.png', res)
def save_images(model_dict,
data_dict,
X_test,
adv_x,
dev_list,
mean,
rd=None,
dr_alg=None):
"""Save <no_of_img> samples as image files in visual_data folder"""
from lib.utils.dr_utils import invert_dr
no_of_img = 1
indices = range(no_of_img)
X_curr = X_test[indices]
channels = data_dict['channels']
atk = model_dict['attack']
dataset = model_dict['dataset']
DR = model_dict['dim_red']
rev = model_dict['rev']
abs_path_v = resolve_path_v(model_dict)
if (rd is not None) and (rev is None):
X_curr = invert_dr(X_curr, dr_alg, DR)
features_per_c = X_curr.shape[-1]
height = int(np.sqrt(features_per_c))
width = height
X_curr_rev = X_curr.reshape((no_of_img, channels, height, width))
elif (rd is None) or ((rd is not None) and (rev is not None)):
height = data_dict['height']
width = data_dict['width']
if channels == 1:
dev_count = 0
for dev_mag in dev_list:
adv_curr = adv_x[indices, :, dev_count]
if (rd is not None) and (rev is None):
adv_x_rev = invert_dr(adv_curr, dr_alg, DR)
adv_x_rev = adv_x_rev.reshape(
(no_of_img, channels, height, width))
for i in indices:
adv = adv_x_rev[i].reshape((height, width))
orig = X_curr_rev[i].reshape((height, width))
adv += mean
orig += mean
img.imsave(
abs_path_v +
'{}_{}_{}_{}_mag{}.png'.format(atk, i, DR, d, dev_mag),
adv * 255,
vmin=0,
vmax=255,
cmap='gray')
img.imsave(abs_path_v +
'{}_{}_{}_orig.png'.format(i, DR, rd),
orig * 255,
vmin=0,
vmax=255,
cmap='gray')
elif (rd is None) or (rev is not None):
adv_x_curr = adv_x[indices, :, dev_count]
for i in indices:
adv = adv_x_curr[i].reshape((height, width))
orig = (X_curr[i] + mean[0]).reshape((height, width))
mean_arr = mean[0].reshape((height, width))
adv += mean[0]
if rd is not None:
fname = abs_path_v + ' {}_{}_{}_rev_{}'.format(
atk, i, DR, rd)
elif rd is None:
fname = abs_path_v + '{}_{}'.format(atk, i)
img.imsave(fname + '_mag{}.png'.format(dev_mag),
adv * 255,
vmin=0,
vmax=255,
cmap='gray')
img.imsave(fname + '_orig.png',
orig * 255,
vmin=0,
vmax=255,
cmap='gray')
dev_count += 1
def main(data_set,
model_type,
latent_dim,
shortcut='True',
num_epoch=100,
log_gamma_decay=0.0,
beta_policy_type='constant'):
data_dir = data_folder(data_set)
if data_dir == '':
print('No such data set named {0}.'.format(data_set))
return
x = load_img_from_folder(data_dir, num_img=50)
if model_type == 'VAE':
variational = True
elif model_type == 'AE':
variational = False
else:
print('No such model type named {0}.'.format(model_type))
return
if shortcut == 'True':
resnet = True
elif shortcut == 'False':
resnet = False
else:
print('Shortcut setting unclear: {0}.'.format(shortcut))
return
if beta_policy_type == 'constant':
beta_policy = constant_beta
elif beta_policy_type == 'linear':
beta_policy = linear_beta
else:
print('No beta policy named {0}.'.format(beta_policy_type))
return
# model = VaeNet(variational, latent_dim=latent_dim, shortcut=resnet, init_log_gamma=-4.0, log_gamma_decay=log_gamma_decay)
# if not os.path.exists('model'):
# os.mkdir('model')
# with tf.Session() as sess:
# sess.run(tf.global_variables_initializer())
# saver = tf.train.Saver()
# writer = tf.summary.FileWriter('graph', sess.graph)
# train_model(model, sess, writer, x, num_epoch, 0.002, 32, beta_policy)
# saver.save(sess, 'model/model.ckpt')
# tf.reset_default_graph()
model = VaeNet(variational,
latent_dim=latent_dim,
shortcut=resnet,
is_train=False)
with tf.Session() as sess:
saver = tf.train.Saver()
saver.restore(
sess,
'./experiment/HYPER_GammaDecay_Policy/Flower_000_constant/model/model.ckpt'
)
samples = generate_sample(model, sess, 500)
if not os.path.exists('samples'):
os.mkdir('samples')
for i in range(500):
mpimg.imsave('samples/' + str(i) + '.jpg', samples[i, :, :, :],
0.0, 1.0)
raw_imgs = draw_boxes(raw_imgs, boxes)
# save found boxes
for box in boxes:
box_list.append(box)
# print(box_list)
# Add heat to each box in box list
heat = add_heat(heat, box_list)
# filter out multiple detections and false positives
tresh_heat = np.copy(heat)
tresh_heat = apply_threshold(tresh_heat, 3)
# visualize the heatmap when displaying
heatmap = np.clip(tresh_heat, 0, 255)
# process labels
labels = label(heatmap)
draw_img = draw_labeled_bboxes(np.copy(img), labels)
print(labels[1], 'cars found in file', fname)
# plot images
mpimg.imsave('./output_images/' + fname[-9:-4] + '_cars_found.jpg',
raw_imgs)
mpimg.imsave('./output_images/' + fname[-9:-4] + '_heat.jpg', heat)
mpimg.imsave('./output_images/' + fname[-9:-4] + '_tresh_heat.jpg',
tresh_heat)
mpimg.imsave('./output_images/' + fname[-9:-4] + '_processed.jpg',
draw_img)
# mpimg.imsave('./output_images/'+fname[-9:-4]+'_windows.jpg', window_img)
def draw_lines(img,
lines,
color=[255, 0, 0],
thickness=6,
output_path="test_images/annotated/"):
"""workflow:
1) examine each individual line returned by hough & determine if it's in left or right lane by its slope
because we are working "upside down" with the array, the left lane will have a negative slope and right positive
2) track extrema
3) compute averages
4) solve for b intercept
5) use extrema to solve for points
6) smooth frames and cache
"""
y_global_min = img.shape[
0] #min will be the "highest" y value, or point down the road away from car
y_max = img.shape[0]
l_slope, r_slope = [], []
l_lane, r_lane = [], []
det_slope = 0.4
hough_lines_img = copy.deepcopy(img)
print "Found {0} lines in the image.".format(len(lines))
for line in lines:
#1
for x1, y1, x2, y2 in line:
cv2.line(hough_lines_img, (x1, y1), (x2, y2), color, thickness)
slope = get_slope(x1, y1, x2, y2)
if slope > det_slope:
r_slope.append(slope)
r_lane.append(line)
elif slope < -det_slope:
l_slope.append(slope)
l_lane.append(line)
#2
y_global_min = min(y1, y2, y_global_min)
mpimg.imsave(output_path + "6a-hough_lines.jpg", hough_lines_img)
# to prevent errors in challenge video from dividing by zero
if ((len(l_lane) == 0) or (len(r_lane) == 0)):
print('no lane detected')
return 1
#3
l_slope_mean = np.mean(l_slope, axis=0)
r_slope_mean = np.mean(r_slope, axis=0)
l_mean = np.mean(np.array(l_lane), axis=0)
r_mean = np.mean(np.array(r_lane), axis=0)
if ((r_slope_mean == 0) or (l_slope_mean == 0)):
print('dividing by zero')
return 1
#4, y=mx+b -> b = y -mx
l_b = l_mean[0][1] - (l_slope_mean * l_mean[0][0])
r_b = r_mean[0][1] - (r_slope_mean * r_mean[0][0])
#5, using y-extrema (#2), b intercept (#4), and slope (#3) solve for x using y=mx+b
# x = (y-b)/m
# these 4 points are our two lines that we will pass to the draw function
l_x1 = int((y_global_min - l_b) / l_slope_mean)
l_x2 = int((y_max - l_b) / l_slope_mean)
r_x1 = int((y_global_min - r_b) / r_slope_mean)
r_x2 = int((y_max - r_b) / r_slope_mean)
#6
if l_x1 > r_x1:
l_x1 = int((l_x1 + r_x1) / 2)
r_x1 = l_x1
l_y1 = int((l_slope_mean * l_x1) + l_b)
r_y1 = int((r_slope_mean * r_x1) + r_b)
l_y2 = int((l_slope_mean * l_x2) + l_b)
r_y2 = int((r_slope_mean * r_x2) + r_b)
else:
l_y1 = y_global_min
l_y2 = y_max
r_y1 = y_global_min
r_y2 = y_max
vertices = np.array([l_x1, l_y1, l_x2, l_y2, r_x1, r_y1, r_x2, r_y2],
dtype=np.int32)
cv2.line(img, (vertices[0], vertices[1]), (vertices[2], vertices[3]),
color, thickness)
cv2.line(img, (vertices[4], vertices[5]), (vertices[6], vertices[7]),
color, thickness)
def process_frame(image, output_path="test_images/annotated/"):
imshape = image.shape
print "imshape =", imshape
# KITCHEN
# lower_left = [0, imshape[0]]
# lower_right = [imshape[1]*9/10, imshape[0]]
# top_left = [0, imshape[0]/4]
# top_right = [imshape[1]*7/8, imshape[0]/4]
# HIGHWAY
# lower_left = [0, imshape[0]]
# lower_right = [imshape[1]*7/9, imshape[0]]
# top_left = [imshape[1]*3/8, imshape[0]*4/10]
# top_right = [imshape[1]*6/9, imshape[0]*4/10]
# BALCANI
# lower_left = [imshape[1]/8, imshape[0]]
# lower_right = [imshape[1]*7/8, imshape[0]]
# top_left = [imshape[1]*3/7, imshape[0]*5/8]
# top_right = [imshape[1]*4/7, imshape[0]*5/8]
# DEFAULT
lower_left = [imshape[1] / 9, imshape[0]]
lower_right = [imshape[1] * 8 / 9, imshape[0]]
top_left = [imshape[1] * 3 / 8, imshape[0] * 6 / 10]
top_right = [imshape[1] * 5 / 8, imshape[0] * 6 / 10]
roi_lines_image = copy.deepcopy(image)
color = [0, 255, 0]
thickness = 3
cv2.line(roi_lines_image, (lower_left[0], lower_left[1]),
(top_left[0], top_left[1]), color, thickness)
cv2.line(roi_lines_image, (top_left[0], top_left[1]),
(top_right[0], top_right[1]), color, thickness)
cv2.line(roi_lines_image, (top_right[0], top_right[1]),
(lower_right[0], lower_right[1]), color, thickness)
cv2.line(roi_lines_image, (lower_right[0], lower_right[1]),
(lower_left[0], lower_left[1]), color, thickness)
mpimg.imsave(output_path + "1a-roi_lines_image.jpg", roi_lines_image)
vertices = [
np.array([lower_left, top_left, top_right, lower_right],
dtype=np.int32)
]
roi_image = region_of_interest(image, vertices)
mpimg.imsave(output_path + "1b-roi_image.jpg", roi_image)
gray_image = grayscale(roi_image)
mpimg.imsave(output_path + "1-grayscale.jpg", gray_image)
img_hsv = cv2.cvtColor(roi_image, cv2.COLOR_RGB2HSV)
mpimg.imsave(output_path + "2-hsv.jpg", img_hsv)
#hsv = [hue, saturation, value]
#more accurate range for yellow since it is not strictly black, white, r, g, or b
lower_yellow = np.array([20, 100, 100], dtype="uint8")
upper_yellow = np.array([30, 255, 255], dtype="uint8")
mask_yellow = cv2.inRange(img_hsv, lower_yellow, upper_yellow)
mpimg.imsave(output_path + "3a-mask_yelow.jpg", mask_yellow)
mask_white = cv2.inRange(gray_image, 180, 255)
mpimg.imsave(output_path + "3b-mask_white.jpg", mask_white)
mask_yw = cv2.bitwise_or(mask_white, mask_yellow)
mask_yw_image = cv2.bitwise_and(gray_image, mask_yw)
mpimg.imsave(output_path + "3c-mask_yellow_white.jpg", mask_yw_image)
kernel_size = 5
gauss_gray = gaussian_blur(mask_yw_image, kernel_size)
mpimg.imsave(output_path + "4-gauss_gray.jpg", gauss_gray)
#same as quiz values
low_threshold = 50
high_threshold = 150
canny_edges = canny(gauss_gray, low_threshold, high_threshold)
mpimg.imsave(output_path + "5-canny_edges.jpg", canny_edges)
#rho and theta are the distance and angular resolution of the grid in Hough space
#same values as quiz
rho = 2
theta = np.pi / 180
#threshold is minimum number of intersections in a grid for candidate line to go to output
threshold = 30
min_line_len = 50
max_line_gap = 100
line_image = hough_lines(canny_edges,
rho,
theta,
threshold,
min_line_len,
max_line_gap,
output_path=output_path)
mpimg.imsave(output_path + "6b-hough_lines.jpg", line_image)
result = weighted_img(line_image, image, alpha=0.8, beta=1., gamma=0.)
return result
return samples
if __name__ == "__main__":
dbTrain = MyDatabase("database/train", "database/train/data_train.csv")
print("Train db length: ", len(dbTrain))
edge = Edge()
dbTest = MyDatabase("database/test", "database/test/data_test.csv")
print("Test db length: ", len(dbTest))
# check shape
assert edge_kernels.shape == (5, 2, 2)
# evaluate database
APs, res = myevaluate(dbTrain,
dbTest,
edge.make_samples,
depth=depth,
d_type="d1")
# add pictures in prediction folder under the predicted class folder
path = "/Users/johanncarfantan/Documents/ENSSAT/IMR3/AnalyseDimages/Partie 1/predictions/"
for i in range(len(dbTest)):
saveName = path + res[i] + "/" + dbTest.data.img[i].split('/')[-1]
bid = imageio.imread(dbTest.data.img[i])
if not os.path.exists(path + res[i]):
os.makedirs(path + res[i])
mpimg.imsave(saveName, bid / 255.)
im = mplimg.imread(imgPath)
# Do rectification in two stages
imA, HA = rectifyAffineF(im, nLinePairs)
plt.close('all')
imM, HM = rectifyMetricF(imA, nLinePairs)
# Translate and scale HM * HA to the image
H = translateAndScaleHToImage(np.dot(HM, HA), imA.shape)
# Apply translated and scaled version of HM * HA
imM = myApplyH(im, H) # Chain the two transforms
plt.close('all')
# Show result
imshow(np.concatenate((im, imA, imM), axis=1))
axis('off')
plt.suptitle(
'Original (left), Affine rectified (middle), Metric rectification (right)')
margin = 0.05
plt.subplots_adjust(left=margin,
right=1.0 - margin,
top=1.0 - margin,
bottom=margin)
show()
# Save output
imgPathRect = fileparts[0] + "_rect" + fileparts[1]
print("Saving output to %s..." % imgPathRect)
mplimg.imsave(imgPathRect, cropOuterRegion(imM))
