image = Image.open(image_dir + image_list)

input_shape = (image_size, image_size)

image = image.resize(input_shape)

image = np.array(image)

# image = image / 127.5

# image = image - 1

image = image / 255.

'''用训练好的权重生成submit文件'''

model = build_model(weights=None, model_name='resnet50')

model.load_weights(weights_path)

print('success creat inception_resnet_v2 model!!!')

'''处理测试图片，生成batch'''

image_list = os.listdir(image_dir)

try:

index = image_list.index('.DS_Store')

image_list = image_list[:index] + image_list[index+1:]

except:

image_list = image_list

num = len(image_list)

r = num % batch_size #15

n = num / batch_size #309

i = 0

batch = 0

preds = []

for a in range(n):

image_batch = []

for b in range(batch_size):

image = get_random_data(image_list[i], image_size, image_dir, is_aug)

image = np.expand_dims(image, axis=0)

image_batch.append(image)

i = i + 1

x = np.concatenate([image for image in image_batch])

y = model.predict_on_batch(x)

#修改这里就可以了

#pred = np.argmax(y, axis = 1) # for top 1

for m in range(y.shape[0]):

pred = np.argsort(y)[m][::-1][:3]

pred = pred.tolist()

preds.append(pred)

batch = batch + 1

print('process %d batch!!' % batch)

if r != 0:

image_batch = []

for c in range(r):

image = get_random_data(image_list[i], image_size, image_dir, is_aug)

image = np.expand_dims(image, axis=0)

image_batch.append(image)

i = i + 1

x = np.concatenate([image for image in image_batch])

y = model.predict_on_batch(x)

#pred = np.argmax(y, axis = 1) #

for l in range(y.shape[0]):

pred = np.argsort(y)[l][::-1][:3]

pred = pred.tolist()

preds.append(pred)

print('process remains!!')

submit = []

for k in range(len(image_list)):

item = {}

item['image_id'] = image_list[k]

item['label_id'] = preds[k]

submit.append(item)

with open(submit_file, 'w') as f:

json.dump(submit, f)

# load submit result and reference result

with open(submit_file, 'r') as file1:

submit_data = json.load(file1)

with open(reference_file, 'r') as file1:

ref_data = json.load(file1)

if len(submit_data) != len(ref_data):

result['warning'].append('Inconsistent number of images between submission and reference data \n')

submit_dict = {}

ref_dict = {}

for item in submit_data:

submit_dict[item['image_id']] = item['label_id']

for item in ref_data:

ref_dict[item['image_id']] = int(item['label_id'])

# eval accuracy

right_count_top3 = 0

right_count_top1 = 0

for (key, value) in ref_dict.items():

if key not in set(submit_dict.keys()):

result['warning'].append('lacking image %s in your submission file \n' % key)

print('warnning: lacking image %s in your submission file' % key)

continue

# if value == submit_dict[key]:

if value in submit_dict[key][:3]:

right_count_top3 += 1

if value == submit_dict[key][0]:

right_count_top1 += 1

result['top3'] = str(float(right_count_top3)/max(len(ref_dict), 1e-5))

result['top1'] = str(float(right_count_top1)/max(len(ref_dict), 1e-5))

if model_name == 'resnet50':

batch_size = 32

image_size = 224

if model_name == 'inception_resnet_v2':

batch_size = 16

image_size = 299

if ucloud:

image_dir_testA = '/data/data/ai_challenger_scene_test_a_20180103/scene_test_a_images_20180103/'

image_dir_testB = '/data/data/ai_challenger_scene_test_b_20180103/scene_test_b_images_20180103/'

weights_path = '/data/code/imagenet_models/scene-output-trained_weights_all_layers_1.h5'

submit_file_testA = '/data/output/testA_submit.json'

submit_file_testB = '/data/output/testB_submit.json'

reference_file_testA = '/data/data/ai_challenger_scene_test_a_20180103/scene_test_a_annotations_20180103.json'

reference_file_testB = '/data/data/ai_challenger_scene_test_b_20180103/scene_test_b_annotations_20180103.json'

testA_zip = '/data/data/ai_challenger_scene_test_a_20180103.zip'

testB_zip = '/data/data/ai_challenger_scene_test_b_20180103.zip'

zip_output = '/data/data/'

unzip(testA_zip, zip_output)

unzip(testB_zip, zip_output)

creat_submit(image_dir_testA, submit_file_testA, image_size, batch_size, weights_path)

creat_submit(image_dir_testB, submit_file_testB, image_size, batch_size, weights_path)

result_A = {'error': [], 'warning': [], 'top3': None, 'top1': None}

submit_dict_A, ref_dict_A, result_A = __load_data(submit_file_testA, reference_file_testA, result_A)

result_A = __eval_result(submit_dict_A, ref_dict_A, result_A)

result_B = {'error': [], 'warning': [], 'top3': None, 'top1': None}

submit_dict_B, ref_dict_B, result_B = __load_data(submit_file_testB, reference_file_testB, result_B)

result_B = __eval_result(submit_dict_B, ref_dict_B, result_B)

print 'testA result =', result_A

print 'testB result =', result_B

else:

#image_dir = '/Users/xiang/Desktop/classification/test/'

image_dir = '/Users/xiang/Desktop/classification/ai_challenger_scene_test_a_20180103/scene_test_a_images_20180103/'

weights_path = 'imagenet_models/scene-output-res50ep008-loss0.981-val_loss0.779-val_acc0.767.h5'

submit_file = 'submit/testA_res50.json'

reference_file = '/Users/xiang/Desktop/classification/ai_challenger_scene_test_a_20180103/scene_test_a_annotations_20180103.json'

creat_submit(image_dir, submit_file, image_size, batch_size, weights_path)

result = {'error': [], 'warning': [], 'top3': None, 'top1': None}

submit_dict, ref_dict, result = __load_data(submit_file, reference_file, result)

result = __eval_result(submit_dict, ref_dict, result)