Compute Items SIFT Features

In [ ]:

    

import json
from utils import load_items

with open('parameters.json', 'r') as infile:
    params = json.load(infile)

RESIZE_X = params['resize']['x']
RESIZE_Y = params['resize']['y']
ITEM_FOLDER = params['item_folder']

items = load_items(ITEM_FOLDER)


    

In [ ]:

    

import cv2, glob
from utils import compute_sift, imread_rgb, imread_gray, pack_keypoint, write_features_to_file

def worker(item):
    folder = ITEM_FOLDER + '/' + item + '/'
    files = glob.glob(folder + '*.png')
    for filename in files:
        image_RGB = imread_rgb(filename)
        if not image_RGB is None:
            image_RGB = cv2.resize(image_RGB,(RESIZE_X,RESIZE_Y))
            file_mask = filename[:-4] + '_mask.pgm'
            mask = imread_gray(file_mask)
            (kp, des) = compute_sift(image_RGB, mask)
            kp, des = pack_keypoint(kp, des) 
            write_features_to_file(filename[:-4] + '_sift.npy', kp, des)


    

In [ ]:

    

%%time
from multiprocessing import Pool

print('Computing SIFTs of images resized to %d x %d' % (RESIZE_X,RESIZE_Y))
pool_size = 6
pool = Pool(pool_size)
result = []
for item in items:
    result.append( pool.apply_async(worker, (item,)) )
pool.close()
pool.join()
for r in result:
     r.get()


    

In [ ]:

    

import glob
from utils import read_features_from_file, unpack_keypoint

item_view = []
num_sift = []
for item in items:
    folder = ITEM_FOLDER + '/' + item + '/'
    files = glob.glob(folder + '*_sift.npy')
    for filename in files:
        kp, des = read_features_from_file(filename)
        kp, des = unpack_keypoint(kp, des)
        item_view.append(filename)
        num_sift.append(len(des))


    

In [ ]:

    

from matplotlib import pyplot as plt
%matplotlib inline
plt.hist(num_sift,bins=60);


    

In [ ]:

    

[(ns, str(iv.split('/')[-1][:-9])) for ns, iv in sorted(zip(num_sift,item_view), reverse=True) if ns>2000]


    

In [ ]:

    

[(ns, str(iv.split('/')[-1][:-9])) for ns, iv in sorted(zip(num_sift,item_view), reverse=True) if ns<50]


    

In [ ]:

    

from matplotlib import pyplot as plt
%matplotlib inline


    

In [ ]:

    

import cv2
from ipywidgets import interact
from utils import imread_rgb, read_features_from_file, unpack_keypoint

def load_and_plot(item,view):
    try:
        prefix = ITEM_FOLDER + '/' + item + '/' + item + '_' + view
        filename = prefix + '_sift.npy'
        kp, des = read_features_from_file(filename)
        kp, des = unpack_keypoint(kp, des)
        filename = prefix + '.png'
        image_RGB = imread_rgb(filename)
        image_RGB = cv2.resize(image_RGB,(RESIZE_X,RESIZE_Y))
        kp_img = cv2.drawKeypoints(image_RGB,kp,color=(0,255,0))
        print('%d features detected' % len(kp))
        plt.imshow(kp_img); plt.axis('off');
    except (IOError, OSError):
        print('File not found')


    

In [ ]:

    

views = ['top_01','top-side_01','top-side_02','bottom_01','bottom-side_01','bottom-side_02']
interact(load_and_plot,item=items,view=views);


    

In [ ]:

    

for item in items:
    for view in views:
        print(item + '_' + view)
        load_and_plot(item,view)
        plt.show()


    

In [ ]:

    

from matplotlib import pyplot as plt
%matplotlib inline


    

In [ ]:

    

import cv2, numpy as np
from ipywidgets import interact
from utils import compute_sift, imread_rgb, imread_gray

def compute_and_plot(item,view):
    prefix = ITEM_FOLDER + '/' + item + '/' + item + '_' + view
    filename = prefix + '.png'
    image_RGB = imread_rgb(filename)
    if not image_RGB is None:
        image_RGB = cv2.resize(image_RGB,(RESIZE_X,RESIZE_Y))
        filename = prefix + '_mask.pgm'
        mask = imread_gray(filename)
        (kp, des) = compute_sift(image_RGB, mask)
        kp_img = cv2.drawKeypoints(image_RGB,kp,color=(0,255,0))
        print('%d features detected' % len(kp))
        plt.imshow(kp_img); plt.axis('off');


    

In [ ]:

    

views = ['top_01','top-side_01','top-side_02','bottom_01','bottom-side_01','bottom-side_02']
interact(compute_and_plot,item=items,view=views);


    

In [ ]:

    

for item in items:
    for view in views:
        print(item + '_' + view)
        plot(item,view)
        plt.show()