summary.add_summary(summary_op_val, count)
# save ckpt files
if count > 0 and count % args.ckpt_interval == 0:
filename = 'InsightFace_iter_{:d}'.format(count) + '.ckpt'
filename = os.path.join(args.ckpt_path, filename)
saver.save(sess, filename)
# validate
if count > 0 and count % args.validate_interval == 0:
feed_dict_test = {dropout_rate: 1.0}
feed_dict_test.update(tl.utils.dict_to_one(net.all_drop))
results = ver_test(ver_list=ver_list,
ver_name_list=ver_name_list,
nbatch=count,
sess=sess,
embedding_tensor=embedding_tensor,
batch_size=args.batch_size,
feed_dict=feed_dict_test,
input_placeholder=images)
print('test accuracy is: ', str(results[0]))
total_accuracy[str(count)] = results[0]
log_file.write('########' * 10 + '\n')
log_file.write(','.join(list(total_accuracy.keys())) +
'\n')
log_file.write(','.join(
[str(val)
for val in list(total_accuracy.values())]) + '\n')
log_file.flush()
if max(results) > 0.996:
print('best accuracy is %.5f' % max(results))
filename = 'InsightFace_iter_best_{:d}'.format(
ver_name_list.append(opt.dataset_name)
images = tf.placeholder(tf.float32, [None, 112, 112, 3], name='image_inputs')
labels = tf.placeholder(tf.int64, [None, ], name='labels_inputs')
w_init_method = tf.contrib.layers.xavier_initializer(uniform=False)
num_classes = opt.num_classes
emb = resnet50(images, is_training=True)
emb = tf.contrib.layers.flatten(emb)
logit = arcface_loss(embedding=emb, labels=labels, w_init=w_init_method, out_num=num_classes)
# inference_loss = tf.reduce_mean(tf.nn.sparse_softmax_cross_entropy_with_logits(logits=logit, labels=labels))
var_list = tf.trainable_variables()
g_list = tf.global_variables()
bn_moving_vars = [g for g in g_list if 'moving_mean' in g.name]
bn_moving_vars += [g for g in g_list if 'moving_variance' in g.name]
var_list += bn_moving_vars
config = tf.ConfigProto()
config.allow_soft_placement = True
config.gpu_options.allow_growth = True
counter = 0
saver = tf.train.Saver(var_list=var_list)
with tf.Session(config=config) as sess:
saver.restore(sess, opt.face_ckpt)
feed_dict = {}
results = ver_test(ver_list=ver_list, ver_name_list=ver_name_list, nbatch=0, sess=sess, embedding_tensor=emb,
batch_size=32, feed_dict=feed_dict, input_placeholder=images)
print(results)
'loss: [%.2f, %.2f], training accuracy is %.6f, time %.3f samples/sec' %
(i, count, total_loss_gpu_1, total_loss_gpu_2, inference_loss_val_gpu_1, inference_loss_val_gpu_2,
wd_loss_val_gpu_1, wd_loss_val_gpu_2, acc_val, pre_sec))
count += 1
# save summary
if count > 0 and count % args.summary_interval == 0:
feed_dict = {images: images_train, labels: labels_train, dropout_rate: 0.4}
summary_op_val = sess.run(summary_op, feed_dict=feed_dict)
summary.add_summary(summary_op_val, count)
# save ckpt files
if count > 0 and count % args.ckpt_interval == 0:
filename = 'InsightFace_iter_{:d}'.format(count) + '.ckpt'
filename = os.path.join(args.ckpt_path, filename)
saver.save(sess, filename)
# # validate
if count >= 0 and count % args.validate_interval == 0:
feed_dict_test ={dropout_rate: 1.0}
results = ver_test(ver_list=ver_list, ver_name_list=ver_name_list, nbatch=count, sess=sess,
embedding_tensor=embedding_tensor, batch_size=args.batch_size//args.num_gpus, feed_dict=feed_dict_test,
input_placeholder=images_test)
if max(results) > 0.998:
print('best accuracy is %.5f' % max(results))
filename = 'InsightFace_iter_best_{:d}'.format(count) + '.ckpt'
filename = os.path.join(args.ckpt_path, filename)
saver.save(sess, filename)
except tf.errors.OutOfRangeError:
print("End of epoch %d" % i)
break
summary_op_val = sess.run(summary_op, feed_dict=feed_dict)
summary.add_summary(summary_op_val, count)
# save ckpt files
if count > 0 and count % ckpt_count_interval == 0:
filename = 'InsightFace_iter_' + str(count) + '-' + str(
i) + '.ckpt'
filename = os.path.join(ckpt_path, filename)
saver.save(sess, filename)
# # validate
if count >= 0 and count % validate_interval == 0:
feed_dict_test = {dropout_rate: 1.0}
results = ver_test(ver_list=ver_list,
ver_name_list=ver_name_list,
nbatch=count,
sess=sess,
embedding_tensor=embedding_tensor,
batch_size=batch_size // len(num_gpus),
feed_dict=feed_dict_test,
input_placeholder=images_test
) # MGPU 增加////len(num_gpus)
if max(results) > 0.998:
print('best accuracy is %.5f' % max(results))
filename = 'InsightFace_iter_best_{:d}'.format(
count) + '.ckpt'
filename = os.path.join(ckpt_path, filename)
saver.save(sess, filename)
except tf.errors.OutOfRangeError:
print("End of epoch %d" % i)
break
