In [1]:

    

%load_ext autoreload
%autoreload 1


    

In [2]:

    

import sys
sys.path.append('src')
from poisson_disk import PoissonDiskSampler
%aimport settings
import settings
%aimport cnn_utils
import cnn_utils
%aimport geometry_utils
import geometry_utils
%aimport dataset_utils
import dataset_utils

import numpy as np
import cv2
import matplotlib
import matplotlib.pylab as plt
sys.path.append(settings.CAFFE_PYTHON_PATH)
import caffe
%matplotlib inline


    

In [3]:

    

plt.rcParams['figure.figsize'] = (10, 10)
plt.rcParams['image.interpolation'] = 'nearest'
plt.rcParams['image.cmap'] = 'gray'


    

In [4]:

    

width = 480
height = 320
radius = 50.
pds = PoissonDiskSampler(width, height, radius, k = 100)
samples = pds.get_sample()
print 'samples size: ', len(samples)
xs = [int(s[0]) for s in samples]
ys = [int(s[1]) for s in samples]
plt.plot(xs, ys, 'o')
plt.xlim([0,width])
plt.ylim([0,height])
plt.show()


    

    




samples size:  42






    

In [5]:

    

dh = cnn_utils.DeepHelper(interpolation=cv2.INTER_NEAREST)


    

In [6]:

    

img = caffe.io.load_image('/home/ipl/datasets/CUB-200-2011/CUB_200_2011/CUB_200_2011/images/084.Red_legged_Kittiwake/Red_Legged_Kittiwake_0062_795434.jpg')


    

In [7]:

    

img.shape


    

    
Out[7]:





(299, 400, 3)

In [8]:

    

plt.imshow(img)


    

    
Out[8]:





<matplotlib.image.AxesImage at 0x7f799b8d4e50>






    

In [9]:

    

dh.init_with_image(img)


    

In [10]:

    

for k in dh.feats.keys():
    print k, dh.feats[k].shape


    

    




conv3 (299, 400, 384)
conv2 (299, 400, 256)
conv1 (299, 400, 96)
conv5 (299, 400, 256)
conv4 (299, 400, 384)

In [11]:

    

plt.matshow(dh.feats['conv1'][:, :, 65])


    

    
Out[11]:





<matplotlib.image.AxesImage at 0x7f799b78b450>






    

In [12]:

    

dh.ffeats.shape


    

    
Out[12]:





(299, 400, 1376)

In [13]:

    

box = geometry_utils.Box(200, 270, 200, 300)
new_img = box.draw_box(img)


    

In [14]:

    

points = box.generate_points_inside(policy='poisson_disk', param=10)
print points.shape
fig = plt.figure(figsize=(10, 10))
ax = fig.add_subplot(111)

geometry_utils.draw_points(points, ax, color='blue')
ax.imshow(new_img)
plt.show()


    

    




(49, 2)






    

In [15]:

    

cub = dataset_utils.CUB_200_2011()
cub_annotations = dataset_utils.BerkeleyAnnotaionHelper(cub)


    

In [16]:

    

imgs_addr = cub.image_addrs()
imgs_bbox = cub.bboxes()


    

In [17]:

    

img_id = 1080

img = caffe.io.load_image(imgs_addr[img_id])
bbox = imgs_bbox[img_id]
head_box = cub_annotations.annotation(img_id, 'head')
body_box = cub_annotations.annotation(img_id, 'body')
plt.imshow(body_box.draw_box(head_box.draw_box(bbox.draw_box(img, (1, 1, 0)), (0, 1, 1)), (1, 0, 1)))


    

    
Out[17]:





<matplotlib.image.AxesImage at 0x7f797b401550>






    

In [18]:

    

IDtrain, IDtest = cub.train_test_id()


    

In [19]:

    

IDtrain.shape, IDtest.shape


    

    
Out[19]:





((5994,), (5794,))

In [20]:

    

IDtrain[:10]


    

    
Out[20]:





array([ 2,  4,  5,  7,  8,  9, 11, 15, 18, 20])

In [21]:

    

plt.imshow(head_box.get_sub_image(img))


    

    
Out[21]:





<matplotlib.image.AxesImage at 0x7f797ad7ac10>






    

In [ ]: