def masking_graph_cost(sig_mask_op):
with tf.name_scope('loss'):
mask_mean = tf.reduce_mean(sig_mask_op, name='mask_loss')
feat_map = presoftmax()
truth_feat_map = tf.placeholder(dtype=tf.float32,
name='target_feat_map')
weight = tf.placeholder(dtype=tf.float32, name='loss_weight')
alpha = tf.placeholder(dtype=tf.float32, name='alpha')
beta = tf.placeholder(dtype=tf.float32, name='beta')
feat_map_diff = (truth_feat_map - feat_map)
feat_map_loss = tf.reduce_mean(tf.square(feat_map_diff) * weight,
name='feat_loss')
# target_softmax = output_label_tensor(label)
feat_map_loss = alpha * feat_map_loss
# total_loss = tf.add(feat_map_loss, total_var(sig_mask_op)*0.5 + beta * mask_mean, name='total_loss')
total_loss = tf.add(feat_map_loss, beta * mask_mean, name='total_loss')
return total_loss, feat_map, (feat_map_loss, mask_mean)
def masking_graph_cost(sig_mask_op):
with tf.name_scope('loss'):
mask_mean = tf.reduce_mean(sig_mask_op, name='mask_loss')
# feat_map = mid_volume()
# feat_map = last_volume()
feat_map = presoftmax()
truth_feat_map = tf.placeholder(dtype=tf.float32,
name='target_feat_map')
alpha = tf.placeholder(dtype=tf.float32, name='alpha')
beta = tf.placeholder(dtype=tf.float32, name='beta')
feat_map_diff = (truth_feat_map - feat_map)
feat_map_loss = tf.reduce_mean(tf.square(feat_map_diff),
name='feat_loss')
feat_map_loss = alpha * feat_map_loss
total_loss = tf.add(feat_map_loss, beta * mask_mean, name='total_loss')
return total_loss, feat_map, (feat_map_loss, mask_mean)
imgs = (2 * imgs / 255) - 1
tf.reset_default_graph()
cfg = tf.ConfigProto()
cfg.gpu_options.allow_growth = True
sess = tf.Session(config=cfg)
with DeepExplain(session=sess, graph=sess.graph) as de:
load_model(model_name,
tf.placeholder(tf.float32, [None, 224, 224, 3], 'input'), True)
labels, ids = top_labels_and_ids(sess, imgs)
# print(labels, ids)
input_n = input_node()
output_n = presoftmax()
target_n = tf.reduce_max(output_n, 1)
attributions = {}
BATCH_SIZE = 10
for i in range(0, len(imgs), BATCH_SIZE):
print('iter %s' % i)
batch = imgs[i:i + BATCH_SIZE]
batch_filenames = img_filenames[i:i + BATCH_SIZE]
# attributions['saliency'] = de.explain('saliency', target_n, input_n, batch)
# attributions['intgrad'] = de.explain('intgrad', target_n, input_n, batch)
# attributions['elrp'] = de.explain('elrp', target_n, input_n, batch)
# attributions['deeplift'] = de.explain('deeplift', target_n, input_n, batch)
attributions['occlusion'] = de.explain('occlusion',
target_n,
input_n,
def eval_img(sess, img, target_id):
return run_network(sess, presoftmax(), [img])[0][target_id]