def data_unit(net, file_name):
n, c, h, w = net.blobs['data'].data.shape
f = open(file_name+'.txt', 'w')
plt.subplot(131)
vu.visualize_one_channel_images(net.blobs['data'].data.reshape(n, h, w))
plt.subplot(132)
vu.visualize_one_channel_images(net.blobs['st_output'].data.reshape(n, h, w))
plt.subplot(133)
acc = np.zeros((n, h, w, 3))
gt_label = net.blobs['label'].data
est_label = np.argmax(net.blobs['class'].data, axis=1)
err = (est_label <> gt_label)
ind = np.array(range(n))[err]
for i in ind:
x = i/ceil(sqrt(n))
y = i%ceil(sqrt(n))
f.write('Digit at (%d, %d) should be %d, but is classified as %d\n'%(x, y, gt_label[i], est_label[i]))
acc[i] = np.ones((h, w, 3))
plt.imshow(vu.vis_grid(acc))
plt.gca().axis('off')
plt.savefig(file_name+'.jpg', dpi = 100)
plt.close()
def data_unit(net, file_name):
n, c, h, w = net.blobs['data'].data.shape
f = open(file_name+'.txt', 'w')
plt.subplot(121)
plt.title('Original Image')
plt.axis('off')
vu.vis_square(net.blobs['downsampled_data'].data.transpose(0, 2, 3, 1))
plt.subplot(122)
plt.axis('off')
plt.title('Correctness')
acc = np.zeros((n, h, w, 3))
gt_label = net.blobs['label'].data
est_label = np.argmax(net.blobs['final/res'].data, axis=1)
err = (est_label <> gt_label)
ind = np.array(range(n))[err]
for i in ind:
x = i/ceil(sqrt(n))
y = i%ceil(sqrt(n))
f.write('Bird at (%d, %d) should be %d, but is classified as %d\n'%(x, y, gt_label[i], est_label[i]))
acc[i] = np.ones((h, w, 3))
plt.imshow(vu.vis_grid(acc))
plt.gca().axis('off')
plt.savefig(file_name+'.jpg', dpi = 1000)
plt.close()
def data_unit(net, file_name):
n, c, h, w = net.blobs['data'].data.shape
f = open(file_name + '.txt', 'w')
plt.subplot(121)
plt.title('Original Image')
plt.axis('off')
vu.vis_square(net.blobs['downsampled_data'].data.transpose(0, 2, 3, 1))
plt.subplot(122)
plt.axis('off')
plt.title('Correctness')
acc = np.zeros((n, h, w, 3))
gt_label = net.blobs['label'].data
est_label = np.argmax(net.blobs['final/res'].data, axis=1)
err = (est_label <> gt_label)
ind = np.array(range(n))[err]
for i in ind:
x = i / ceil(sqrt(n))
y = i % ceil(sqrt(n))
f.write('Bird at (%d, %d) should be %d, but is classified as %d\n' %
(x, y, gt_label[i], est_label[i]))
acc[i] = np.ones((h, w, 3))
plt.imshow(vu.vis_grid(acc))
plt.gca().axis('off')
plt.savefig(file_name + '.jpg', dpi=1000)
plt.close()
def data_unit(net, file_name):
n, c, h, w = net.blobs["data"].data.shape
plt.subplot(131)
plt.title("Original Image")
plt.axis("off")
vu.visualize_one_channel_images(net.blobs["data"].data.reshape(n, h, w))
plt.subplot(132)
plt.title("ST Output")
plt.axis("off")
vu.visualize_one_channel_images(net.blobs["st_output"].data.reshape(n, h, w))
plt.subplot(133)
plt.axis("off")
plt.title("Correctness")
acc = np.zeros((n, h, w, 3))
gt_label = net.blobs["label"].data
est_label = np.argmax(net.blobs["class"].data, axis=1)
err = est_label <> gt_label
ind = np.array(range(n))[err]
for i in ind:
acc[i] = np.ones((h, w, 3))
plt.imshow(vu.vis_grid(acc))
plt.gca().axis("off")
plt.savefig(file_name + ".jpg", dpi=1000)
plt.close()
def data_unit(net, file_name):
n, c, h, w = net.blobs['data'].data.shape
plt.subplot(131)
plt.title('Original Image')
plt.axis('off')
vu.visualize_one_channel_images(net.blobs['data'].data.reshape(n, h, w))
plt.subplot(132)
plt.title('ST Output')
plt.axis('off')
vu.visualize_one_channel_images(net.blobs['st_output'].data.reshape(n, h, w))
plt.subplot(133)
plt.axis('off')
plt.title('Correctness')
acc = np.zeros((n, h, w, 3))
gt_label = net.blobs['label'].data
est_label = np.argmax(net.blobs['class'].data, axis=1)
err = (est_label <> gt_label)
ind = np.array(range(n))[err]
for i in ind:
acc[i] = np.ones((h, w, 3))
plt.imshow(vu.vis_grid(acc))
plt.gca().axis('off')
plt.savefig(file_name+'.jpg', dpi = 1000)
plt.close()
def data_unit(net, file_name):
n, c, h, w = net.blobs['data'].data.shape
plt.subplot(131)
plt.title('Original Image')
plt.axis('off')
vu.vis_square(net.blobs['data'].data.transpose(0, 2, 3, 1))
plt.subplot(132)
plt.title('Mask_output')
plt.axis('off')
vu.vis_square(net.blobs['mask_output'].data.transpose(0, 2, 3, 1))
plt.subplot(133)
plt.axis('off')
plt.title('Correctness')
acc = np.zeros((n, h, w, 3))
gt_label = net.blobs['label'].data
est_label = np.argmax(net.blobs['loss3/classifier'].data, axis=1)
err = (est_label <> gt_label)
ind = np.array(range(n))[err]
for i in ind:
acc[i] = np.ones((h, w, 3))
plt.imshow(vu.vis_grid(acc))
plt.gca().axis('off')
plt.savefig(file_name+'.jpg', dpi = 1000)
plt.close()
def data_unit(net, file_name):
n, c, h, w = net.blobs["data"].data.shape
f = open(file_name + ".txt", "w")
plt.subplot(221)
plt.title("Original Image")
plt.axis("off")
vu.vis_square(net.blobs["downsampled_data"].data.transpose(0, 2, 3, 1))
plt.subplot(223)
plt.title("Inc1/data")
plt.axis("off")
print net.blobs["st/theta_1"].data
vu.vis_square(net.blobs["inc1/data"].data.transpose(0, 2, 3, 1))
plt.subplot(224)
plt.title("Inc2/data")
plt.axis("off")
print net.blobs["st/theta_2"].data
vu.vis_square(net.blobs["inc2/data"].data.transpose(0, 2, 3, 1))
plt.subplot(222)
plt.axis("off")
plt.title("Correctness")
acc = np.zeros((n, h, w, 3))
gt_label = net.blobs["label"].data
est_label = np.argmax(net.blobs["final/res"].data, axis=1)
err = est_label <> gt_label
ind = np.array(range(n))[err]
for i in ind:
x = i / ceil(sqrt(n))
y = i % ceil(sqrt(n))
f.write("Bird at (%d, %d) should be %d, but is classified as %d\n" % (x, y, gt_label[i], est_label[i]))
acc[i] = np.ones((h, w, 3))
plt.imshow(vu.vis_grid(acc))
plt.gca().axis("off")
plt.savefig(file_name + ".jpg", dpi=1000)
plt.close()
def data_unit(net, file_name):
n, c, h, w = net.blobs['data'].data.shape
plt.subplot(221)
plt.title('Original Image')
plt.axis('off')
vu.vis_square(net.blobs['data'].data.transpose(0, 2, 3, 1))
plt.subplot(223)
plt.title('Inc1/data')
plt.axis('off')
vu.vis_square(net.blobs['inc1/data'].data.transpose(0, 2, 3, 1))
plt.subplot(224)
plt.title('Inc2/data')
plt.axis('off')
vu.vis_square(net.blobs['inc2/data'].data.transpose(0, 2, 3, 1))
plt.subplot(222)
plt.axis('off')
plt.title('Correctness')
acc = np.zeros((n, h, w, 3))
gt_label = net.blobs['label'].data
est_label = np.argmax(net.blobs['final/res'].data, axis=1)
err = (est_label <> gt_label)
ind = np.array(range(n))[err]
for i in ind:
acc[i] = np.ones((h, w, 3))
plt.imshow(vu.vis_grid(acc))
plt.gca().axis('off')
plt.savefig(file_name+'.jpg')
plt.close()