def visualize(relevances, images_tensor=None, name='', y_labels=''):
n, w, h, dim = relevances.shape
heatmaps = []
#import pdb;pdb.set_trace()
if images_tensor is not None:
assert relevances.shape == images_tensor.shape, 'Relevances shape != Images shape'
for h, heat in enumerate(relevances):
if images_tensor is not None:
input_image = images_tensor[h]
maps = render.hm_to_rgb(heat, input_image, scaling=3, sigma=2)
else:
maps = render.hm_to_rgb(heat, scaling=3, sigma=2)
heatmaps.append(maps)
R = np.array(heatmaps)
with tf.name_scope(name):
if (y_labels != ''):
rel_imgs = tf.cast(R, tf.float32).eval()
for i, rel_img in enumerate(rel_imgs):
cv2.putText(rel_img, str(y_labels[i]), (2, 30),
cv2.FONT_HERSHEY_SIMPLEX, 1, 255)
rel_imgs[i] = cv2.cvtColor(
(np.asarray(rel_img, dtype='float32') * 255).astype(
np.uint8), cv2.COLOR_RGB2BGR)
#output to directory:
#cv2.imwrite(os.path.join("heat_maps",str(i)+".jpg"),rel_imgs[i])
return tf.summary.image('relevance', tf.cast(rel_imgs, tf.float32),
n).eval()
else:
img = tf.summary.image('relevance', tf.cast(R, tf.float32), n)
return img.eval()
def visualize(relevances, images_tensor=None, name=''):
n, w, h, dim = relevances.shape
heatmaps = []
#import pdb;pdb.set_trace()
if images_tensor is not None:
assert relevances.shape == images_tensor.shape, 'Relevances shape != Images shape'
for h, heat in enumerate(relevances):
if images_tensor is not None:
input_image = images_tensor[h]
maps = render.hm_to_rgb(heat, input_image, scaling=3, sigma=2)
else:
maps = render.hm_to_rgb(heat, scaling=3, sigma=2)
heatmaps.append(maps)
R = np.array(heatmaps)
with tf.name_scope(name):
img = tf.summary.image('relevance', tf.cast(R, tf.float32), n)
return img.eval()
def visualize(relevances, images_tensor):
n, dim = relevances.shape
heatmap = relevances.reshape([n, 28, 28, 1])
input_images = images_tensor.reshape([n, 28, 28, 1])
heatmaps = []
for h, heat in enumerate(heatmap):
input_image = input_images[h]
maps = render.hm_to_rgb(heat, input_image, scaling=3, sigma=2)
heatmaps.append(maps)
R = np.array(heatmaps)
with tf.name_scope('input_reshape'):
img = tf.summary.image('input', tf.cast(R, tf.float32), n)
return img.eval()
def visualize(relevances,
images_tensor=None,
image_show=False,
image_save=False):
n, w, h, dim = relevances.shape
heatmaps = []
count = 0
#import pdb;pdb.set_trace()
if images_tensor is not None:
assert relevances.shape == images_tensor.shape, 'Relevances shape != Images shape'
for h, heat in enumerate(relevances):
if images_tensor is not None:
input_image = images_tensor[h]
maps = render.hm_to_rgb(heat, input_image, scaling=3, sigma=2)
else:
maps = render.hm_to_rgb(heat, scaling=3, sigma=2)
heatmaps.append(maps)
if image_show:
plt.imshow(maps)
if image_save:
plt.imsave('saved_image/' + str(count), maps, format="png")
count += 1
R = np.array(heatmaps)
def visualize(relevances, images_tensor):
n = FLAGS.batch_size
heatmap = np.sum(relevances.reshape([n, 224, 224, 3]),axis=3)
input_images = images_tensor.reshape([n, 224, 224, 3])
heatmaps = []
for h, heat in enumerate(heatmap):
input_image = input_images[h]
maps = render.hm_to_rgb(heat, input_image, scaling=3, sigma=1,cmap = 'seismic')
heatmaps.append(maps)
imageio.imsave('./result/imgnet/'+FLAGS.relevance_method+'/heatmap' + str(h) + '.jpg', maps,vmax=1,vmin=-1)
imageio.imsave(
'./result/imgnet/' + FLAGS.relevance_method + '/img' + str(h) + '.jpg', render.enlarge_image(input_image,3))
np.save('./result/imgnet/' + FLAGS.relevance_method + '/' + '_heatmap.npy', relevances)
R = np.array(heatmaps)
with tf.name_scope('input_reshape'):
img = tf.summary.image('input', tf.cast(R, tf.float32), n)
return img.eval()
def visualize(relevances, images_tensor):
n = len(relevances)
for j in range(n):
heatmaps = []
for i in range(32):
heatmap = relevances[j, i, :, :, :]
input_images = images_tensor[j, i, :, :, :]
maps = render.hm_to_rgb(np.squeeze(heatmap), scaling=8, sigma=2, cmap='bwr')
img = input_images
if heatmaps == []:
heatmaps = np.expand_dims(maps,0)
imgs = np.expand_dims(img,0)
else:
heatmaps = np.concatenate([heatmaps,np.expand_dims(maps,0)],0)
imgs = np.concatenate([imgs,np.expand_dims(img,0)],0)
imageio.mimsave('./Demo_img/heatmap'+str(j)+'.gif', heatmaps)
R = np.array(heatmaps)
with tf.name_scope('input_reshape'):
img = tf.summary.image('input', tf.cast(R, tf.float32), n)
return img.eval()
