Exemple #1
0
def main():

    # Parse arguments
    parser = argparse.ArgumentParser(description='Classify images in a directory as good or bad.')
    parser.add_argument('images-path', nargs='?', default='.', help='Path to directory of input images.')
    args = vars(parser.parse_args())

    # Select directory
    global if_path
    if_path = args['images-path']

    # Parse in features information
    global feats
    cwd = os.getcwd()
    os.chdir('../suitability/')
    feats = Feats(if_path)
    print('- Loaded features for %d files' % len(feats))

    # Initialise classifier
    global classifier
    classifier = Classifier()
    os.chdir(cwd)

    # Get list of files
    global files
    valid_suffices = ('png', 'jpg', 'jpeg')
    files = [f for f in os.listdir(if_path) if \
             os.path.isfile('%s/%s' % (if_path, f)) and \
             f.lower().endswith(valid_suffices)]
    if len(files) == 0:
        print('No images to read in: %s' % if_path)
        cleanup()

    # Classify each image and keep suitable ones
    global suitability_scores
    suitable = []
    suitability_scores = []
    for fname in files:
        score = classifier.getScore(feats[fname])
        if score > 0:
            suitable.append(fname)
            suitability_scores.append(score)

    print('- Classified images to retain list of suitable images only')
    print('- %d/%d suitable images left' % (len(suitable), len(files)))
    files = suitable

    # Create list of last shown times
    global last_times
    global selected
    last_times = [time.time()]*len(files)
    selected = [False]*len(files)

    # Main loop
    global num
    num = 0
    find_next_image()
    show_next_image()

    signal.signal(signal.SIGINT, cleanup)

    while True:
        # Wait for key for 50ms
        k = cv.waitKey(50)

        # If no more images left to show, exit
        n_left = len([x for x in selected if not x])
        if n_left == 0:
            break

        # Ignore if no key captured
        if k is -1:
            continue

        # If Esc, break loop
        elif k == 27:
            break

        # If any other button, show previous image
        else:
            next_image()
            print '%d images left to show' % n_left

    cleanup()
Exemple #2
0
def get_images(n):

    # Parse in features information
    global file_to_feat
    cwd = os.getcwd()
    os.chdir('../suitability/')
    feats = Feats(if_path)
    print('- Loaded features for %d files' % len(feats))

    # Initialise classifier
    global classifier
    classifier = Classifier()
    os.chdir(cwd)

    # Get list of files
    global files
    valid_suffices = ('png', 'jpg', 'jpeg')
    files = [f for f in os.listdir(if_path) if \
             os.path.isfile('%s/%s' % (if_path, f)) and \
             f.lower().endswith(valid_suffices)]
    if len(files) == 0:
        print('No images to read in: %s' % if_path)
        cleanup()
    random.shuffle(files)

    # Classify each image and keep suitable ones
    suitable = [False]*len(files)
    scores = [0]*len(files)
    for i, fpath in enumerate(files):
        if classifier.predictFeats(feats[fpath]) == 1:
            suitable[i] = True
        scores[i] = classifier.getScore(feats[fpath])

    # Show 10 top suitable images
    N = 7
    _, topfs = zip(*sorted(zip(scores, files), reverse=True)[:N])
    iH = 150
    iW = 200
    H = iH
    W = iW*N
    out = np.ndarray((H, W, 3), dtype=np.uint8)
    out.fill(255)
    for i, fpath in enumerate(topfs):
        I = imread_rotated('%s/%s' % (if_path, fpath))
        h, w, _ = I.shape
        scale = min(float(iW) / w, float(iH) / h)
        I = cv.resize(I, None, fx=scale, fy=scale)
        h, w, _ = I.shape
        x0 = iW*i + iW/2 - w/2
        y0 = iH/2 - h/2
        out[y0:y0+h, x0:x0+w,] = I
    cv.imwrite('top_suitable.jpg', out)

    # Get up to n entries
    #pairs = sorted(zip(suitable, files))
    #good = [(files[i], suitable[i]) for i in range(len(files)) if suitable[i]]
    #bad = [(files[i], suitable[i]) for i in range(len(files)) if not suitable[i]]
    #random.shuffle(good)
    #random.shuffle(bad)
    #n_good = int(n * 0.35)
    #n_bad = n - n_good
    #pairs = bad[:n_bad] + good[:n_good]
    #random.shuffle(pairs)
    #files, suitable = zip(*pairs)
    files = files[:n]
    suitable = suitable[:n]

    files = ['%s/%s' % (if_path, fname) for fname in files]

    # Load images
    imgs = [None]*len(files)
    for i, fpath in enumerate(files):
        imgs[i] = imread_rotated(fpath)

    return files, imgs, suitable