multiwayPython/run_willow.py at main · Multi-View-Image-Matching-Swarthmore/multiwayPython · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
import glob
import numpy as np
import tensorflow as tf
from scipy.misc import imread, imresize
import scipy.io
from myalexnet_forward import network
from myalexnet_forward import run
from hypercol import compute_hypercols
import os, sys
from tqdm import tqdm

classes = ["Car", "Duck", "Face", "Motorbike", "Winebottle"]

RESIZEFACTOR = 1.0
PATH_LIST = glob.glob('./dataset/WILLOW-ObjectClass/*')
print(PATH_LIST)

datasetFilePath = os.getcwd() + "/dataset/WILLOW-ObjectClass/"
print(datasetFilePath)

# for c in classes:
#     files = os.listdir(datasetFilePath + c + "/")
#     for f in files:
#         if f.endswith(".mat"):
#             npFilename = f.replace(".mat", ".npy")
#             npFilepath = os.getcwd() + "/dataset/numpyFiles/" + c + "/"
#             print(datasetFilePath + c + "/" + f)
#             mat = scipy.io.loadmat(datasetFilePath + c + "/" + f)
#             sys.exit()
#             # print(mat)
#             arr = mat['pts_coord']
#             nparr = np.array(arr)
#             # print(nparr.shape)
#             np.save(npFilepath + npFilename, nparr)
#             print("processed", f)

######
for i in tqdm(range(len(PATH_LIST))):
    pname = PATH_LIST[i]
    fnames = glob.glob(pname+'/*.png')
    # load and resize images (will also save them afterwards)
    ims = [imresize(imread(f)[:,:,:3], RESIZEFACTOR).astype('float32') for f in fnames]
    # load all desired pixel positions
    kptsfiles = [f[0:-3]+'mat' for f in fnames]
    kpts = [scipy.io.loadmat(f)['pts_coord'] for f in kptsfiles]
    # resize and convert from x,y to i,j
    kpts = [RESIZEFACTOR*k[[1,0],:] for k in kpts]
    layers = ['c4', 'c5']
    tf.reset_default_graph()
    net = network()

    # we can't send batches because each image has different dimensions
    for im, k, f in zip(ims, kpts, fnames):
        out = run([im], net, layers)
        hc = np.squeeze(compute_hypercols(out, k.T, im.shape[0]))
        hc = (np.transpose(hc)).astype('float32')
        scipy.io.savemat(f + '.hypercols_kpts.mat',
                         {'desc': hc,
                          'frame': k[[1,0], :],
                          'nfeature': hc.shape[1],
                          'img': im.astype('uint8'),
                          'filename': f[len(pname)+1:]
                          })