def train(trainX, trainY): m = SRCNN() m.compile(Adam(lr=0.0003), 'mse') count = 1 while True: m.fit(trainX, trainY, batch_size=128, nb_epoch=5) print("Saving model " + str(count * 5)) m.save(join('./model_' + str(count * 5) + '.h5')) count += 1
def train_batch(aug, trainX, trainY, load_weight=True): print("[INFO] compiling model...") model = SRCNN() if load_weight and os.path.exists('my_model_weights.h5'): print("[INFO] import model weights...") model.load_weights('my_model_weights.h5') try: opt = Adam(lr=INIT_LR, decay=INIT_LR / EPOCHS) # opt = Adam(lr=INIT_LR) model.compile(loss="mean_squared_error", optimizer=opt, metrics=[psnr1]) current_time = datetime.now().strftime("%Y%m%d-%H:%M") filepath = 'train_model/model-ep{epoch:03d}-loss{loss:.3f}-psnr{psnr1:.3f}-' + current_time + '.h5' checkpoint = ModelCheckpoint(filepath, monitor='psnr1', verbose=1, save_best_only=True, mode='max', period=2) print("[INFO] training network...") H = model.fit_generator(aug.flow(trainX, trainY, batch_size=BS), steps_per_epoch=len(trainX) // BS, epochs=EPOCHS, callbacks=[checkpoint]) print("[INFO] training finish") except KeyboardInterrupt: print("[INFO] KeyboardInterrupted...") except Exception as e: print(e) else: plt.figure() N = EPOCHS # plt.plot(np.arange(0, N), H.history["loss"], label="train_loss") # plt.plot(np.arange(0, N), H.history["acc"], label="train_acc") plt.plot(np.arange(0, N), H.history["psnr1"], label="psnr") plt.xlabel("Epoch #") plt.ylabel("psnr") plt.legend(loc="upper left") plt.savefig("plot.png") finally: print("[INFO] serializing network...") model.save("./model.h5") model.save_weights('my_model_weights.h5')
image_size = args.Image_patch_size #image_patch_size stride = args.Stride scale = args.Scale learning_rate = args.lr batch_size = args.BATCH_SIZE epochs = args.Epochs is_training =args.is_training dirname_train = args.dirname_train dirname_test = args.dirname_test if is_training: X_train,Y_train = load_train(image_size = image_size,stride = stride,scale = scale,dirname =dirname_train) srcnn = SRCNN( image_size = image_size, learning_rate=learning_rate) optimizer = Adam(lr = learning_rate) srcnn.compile(optimizer=optimizer, loss='mean_squared_error') history = srcnn.fit(X_train, Y_train, batch_size=batch_size, epochs=epochs, verbose=2,validation_split = 0.1 ) draw_loss_plot(history = history) #Saving trained model_weights in the current workspace . #make a folder named srcnn in your current workspace you are working in.In that folder your weights will be stored. json_string = srcnn.to_json() open(os.path.join('./srcnn/','srcnn_model.json'),'w').write(json_string) srcnn.save_weights(os.path.join('./srcnn/','srcnn_weight.hdf5')) else: dir_list = os.listdir(dirname_test) for img in dir_list: image = cv2.cvtColor(cv2.imread(os.path.join(dirname_test,img)),cv2.COLOR_BGR2GRAY)