def test(self, writer):
if self.use_exif_summary:
exif_start = time.time()
test_queue = self.net.train_runner.get_random_test(
batch_size=self.net.batch_size)
to_print = []
for i, (im_a_batch, im_b_batch,
label_batch) in enumerate(test_queue):
tag = self.net.train_runner.tags[i]
output = self.sess.run(self.net.pred,
feed_dict={
self.net.im_a: im_a_batch,
self.net.im_b: im_b_batch,
self.net.label: label_batch,
self.net.is_training: False
})
tag_acc = 100.0 * (
np.sum(np.round(output[:, i]) == label_batch[:, i]) /
float(self.net.batch_size))
summary = self.sess.run(
self.individual_summary[tag],
feed_dict={self.tag_holder[tag]: tag_acc})
io.add_summary(writer, [summary], self.i)
to_print.append([tag, tag_acc])
io.show(to_print, phase='test', iter=self.i)
print('EXIF test accuracy evaluation took %.2f seconds' %
(time.time() - exif_start))
return
def show(self, writer, phase='train'):
if self.net.use_tf_threading:
summary = self.sess.run(self.summary)
else:
args = self.get_data(self.net.batch_size, phase)
summary = self.sess.run(self.summary, feed_dict=args)
io.add_summary(writer, summary, self.i)
io.show([['Train time', np.mean(list(self.train_timer))]],
phase=phase, iter=self.i)
return