make_submission.py

import os
import cv2
from tqdm import tqdm
import numpy as np
import pandas as pd

from keras.applications.mobilenet import MobileNet
from keras.models import Model
from keras.layers import Activation, GlobalAveragePooling2D, Dense

test_dir = "./DogBreed/test/"
sample_submission_path = "./sample_submission.csv"

base_model = MobileNet(input_shape=(224, 224, 3), include_top=False)
x = base_model.output
x = GlobalAveragePooling2D()(x)
x = Dense(1024, activation='relu')(x)
predictions = Dense(120, activation='softmax')(x)
model = Model(input=base_model.input, output=predictions)
model.load_weights("./models/1_w_01_0.68.hdf5")

df_test = pd.read_csv(sample_submission_path)
# print(df_test.iloc[0])
dog_breeds = list(df_test.columns.values)[1:]
filenames = list(set(df_test.id))

# ger_shepard_path = "./german_shepherd/"
# filenames = os.listdir(ger_shepard_path)

for idx in tqdm(range(len(filenames))):
    filename = df_test.loc[idx,"id"]
    file_path = test_dir + filename + ".jpg"
    # file_path = ger_shepard_path + filename + ".jpg"

    if not os.path.isfile(file_path):
        print("DEBUG: file doest exist: %s", file_path)
        continue

    img = cv2.imread(file_path)
    img_in = cv2.resize(img, (224, 224), interpolation=cv2.INTER_CUBIC)
    img_in = img_in / 255.
    img_in = np.reshape(img_in, (1, 224, 224, 3))

    y = model.predict(img_in)
    # print(filename)
    # idx_max = np.argmax(y[0])
    # print(dog_breeds[idx_max])
    #
    # cv2.imshow("img", img)
    # cv2.waitKey()
    # print(y[0])

    df_test.loc[idx, 1:] = y[0]

print(df_test)
df_test.to_csv("subm_3.csv", index=False)