Fixed image size, fileprefix and other minor changes

.gitignore

@@ -113,8 +113,12 @@ experiments/
 #  option (not recommended) you can uncomment the following to ignore the entire idea folder.
 .idea/
 
-commandline
 */cpkt*
+*.index
+*.pb
+*.data-*
+
+commandline
 *.png
 *.json
 *.txt

README.md

@@ -25,7 +25,7 @@ A SuperResolution training program for creating/training upscaling machine learn
 ### EDSR - Enhanced Deep Residual Networks for Single Image Super-Resolution
 
 ```bash
-python superresolution/SuperResolution.py --batch-size 16 --epochs 10 --image-size 128 128 --model edsr --learning-rate 0.0003 --color-space rgb --loss-fn msa --shuffle-data-set-size 512 --show-psnr --data-set-directory /path_to_training_data/   --output-dir image-super-resolution-result/
+python superresolution/SuperResolution.py --batch-size 16 --epochs 10 --image-size 128 128 --model edsr --learning-rate 0.0003 --decay-rate 0.9 --decay-step 10000 --color-space rgb --loss-fn msa --shuffle-data-set-size 1024 --show-psnr --data-set-directory /path_to_training_data/   --output-dir image-super-resolution-result/
 ```
 
 ![Gangsta Anime EDSR Super Resolution Example from Trained model](https://github.com/voldien/SuperResolution/assets/9608088/1951a0c3-cebb-4ea8-818e-4a04bf28e116)

superresolution/SuperResolution.py

@@ -186,8 +186,10 @@ def run_train_model(args: dict, training_dataset: Dataset, validation_dataset: D
 
 	# Create Input and Output Size
 	# TODO determine size.
+	resolution_upscale_constant: int = 2
 	image_input_size = (
-		int(args.image_size[0] / 2), int(args.image_size[1] / 2), args.color_channels)
+		int(args.image_size[0] / resolution_upscale_constant), int(args.image_size[1] / resolution_upscale_constant),
+		args.color_channels)
 	image_output_size = (
 		args.image_size[0], args.image_size[1], args.color_channels)
 
@@ -246,7 +248,6 @@ def run_train_model(args: dict, training_dataset: Dataset, validation_dataset: D
 		validation_data_ds = configure_dataset_performance(ds=validation_dataset, use_cache=False,
 														   cache_path=None,
 														   shuffle_size=0)
-
 		# Apply data augmentation
 		validation_data_ds = augment_dataset(dataset=validation_data_ds, image_crop_shape=image_output_size)
 
@@ -318,9 +319,15 @@ def run_train_model(args: dict, training_dataset: Dataset, validation_dataset: D
 		example_result_call_back = SaveExampleResultImageCallBack(
 			args.output_dir,
 			non_augmented_dataset_train, args.color_space,
-			nth_batch_sample=args.example_batch, grid_size=args.example_batch_grid_size)
+			nth_batch_sample=args.example_nth_batch, grid_size=args.example_nth_batch_grid_size)
 		training_callbacks.append(example_result_call_back)
 
+		# Debug output of trained data.
+		#training_callbacks.append(SaveExampleResultImageCallBack(
+		#	args.output_dir,
+		#	training_dataset, args.color_space, fileprefix="trainSuperResolution",
+		#	nth_batch_sample=args.example_nth_batch, grid_size=args.example_nth_batch_grid_size))
+
 		composite_train_callback = CompositeImageResultCallBack(
 			dir_path=args.output_dir,
 			name="train",
@@ -396,14 +403,13 @@ def dcsuperresolution_program(vargs=None):
 		parser.add_argument('--output-dir', type=str, dest='output_dir',
 							default=str.format("super-resolution-{0}", date.today().strftime("%b-%d-%Y_%H:%M:%S")),
 							help='Set the output directory that all the models and results will be stored at')
-
 		#
-		parser.add_argument('--example-batch', dest='example_batch', required=False,  # TODO rename
+		parser.add_argument('--example-batch', dest='example_nth_batch', required=False,  # TODO rename
 							type=int,
 							default=1024,
 							help='Set the number of train batches between saving work in progress result.')
 		#
-		parser.add_argument('--example-batch-grid-size', dest='example_batch_grid_size',
+		parser.add_argument('--example-batch-grid-size', dest='example_nth_batch_grid_size',
 							type=int, required=False,
 							default=8, help='Set the grid size of number of example images.')
 
@@ -465,7 +471,7 @@ def dcsuperresolution_program(vargs=None):
 
 		sr_logger.info(str.format("Use RAM Cache: {0}", args.cache_ram))
 
-		sr_logger.info(str.format("Example Batch Grid Size: {0}", args.example_batch_grid_size))
+		sr_logger.info(str.format("Example Batch Grid Size: {0}", args.example_nth_batch_grid_size))
 		sr_logger.info(str.format("Image Training Set: {0}", args.image_size))
 		sr_logger.info(str.format("Learning Rate: {0}", args.learning_rate))
 		sr_logger.info(str.format(

superresolution/core/common.py

@@ -93,9 +93,6 @@ def DefaultArgumentParser() -> argparse.ArgumentParser:
 	#
 	parser.add_argument('--seed', type=int, default=randrange(10000000), dest='seed',
 						help='Set the random seed')
-
-	parser.add_argument('--nr_image_example_generate', type=int, default=16, dest='num_examples_to_generate',
-						help='Number')
 	#
 	parser.add_argument('--color-space', type=str, default="rgb", dest='color_space', choices=['rgb', 'lab'],
 						help='Select Color Space used in the model.')

superresolution/util/dataProcessing.py

@@ -98,21 +98,22 @@ def augment_dataset(dataset: Dataset, image_crop_shape: tuple) -> Dataset:
 	:return: Augmented DataSet
 	"""
 	trainAug = tf.keras.Sequential([
-		# Select random section of Image.
-		tf.keras.layers.RandomCrop(
-			image_crop_shape[0], image_crop_shape[1]),
-		# Flip image around on each axis randomly.
-		layers.RandomFlip("horizontal_and_vertical"),
 		# Random Zoom.
 		layers.RandomZoom(
-			height_factor=(-0.05, 0.05),
-			width_factor=(-0.05, 0.05),
+			height_factor=(-0.1, 0.1),
+			width_factor=(-0.1, 0.1),
 			fill_mode='reflect',
 			interpolation='bilinear'),
 		# Random Rotation.
 		layers.RandomRotation(factor=0.65,
 							  fill_mode='reflect',
-							  interpolation='bilinear')
+							  interpolation='bilinear'),
+		# Select random section of Image.
+		tf.keras.layers.RandomCrop(
+			image_crop_shape[0], image_crop_shape[1]),
+		# Flip image around on each axis randomly.
+		layers.RandomFlip("horizontal_and_vertical"),
+
 	])
 
 	def AgumentFunc(x):
@@ -127,7 +128,7 @@ def AgumentFunc(x):
 	return dataset
 
 
-def dataset_super_resolution(dataset: Dataset, input_size, output_size) -> Dataset:
+def dataset_super_resolution(dataset: Dataset, input_size: tuple, output_size: tuple) -> Dataset:
 	"""
 	Perform Super Resolution Data and Expected Data to Correct Size. For providing
 	the model with corrected sized Data.
@@ -150,6 +151,14 @@ def DownScaleLayer(data):
 				interpolation='bilinear',
 				crop_to_aspect_ratio=False
 			)])
+
+		expectedScale = tf.keras.Sequential([
+			layers.Resizing(
+				output_size[0],
+				output_size[1],
+				interpolation='bilinear',
+				crop_to_aspect_ratio=False
+			)])
 
 		# Create a copy to prevent augmentation be done twice separately.
 		expected = tf.identity(data)
@@ -161,7 +170,11 @@ def DownScaleLayer(data):
 		# Remap from [0, 1] to [-1,1]
 		data = (2.0 * data) - 1.0
 
-		return data, expected
+		expected_data = (expected + 1.0) * 0.5
+		expected_data = expectedScale(expected_data)
+		expected_data = (2.0 * expected_data) - 1.0
+
+		return data, expected_data
 
 	DownScaledDataSet = (
 		dataset

superresolution/util/image.py

@@ -2,9 +2,10 @@
 from util.util import convert_nontensor_color_space
 import tensorflow as tf
 from numpy import asarray
+from tensorflow.python.data import Dataset
 
 
-def show_expect_predicted_result(model: tf.keras.Model, image_batch_dataset, color_space: str, nr_col=6):
+def show_expect_predicted_result(model: tf.keras.Model, image_batch_dataset: Dataset, color_space: str, nr_col=6):
 	batch_iter = iter(image_batch_dataset)
 	data_image_batch, expected_image_batch = batch_iter.next()
 

superresolution/util/trainingcallback.py

@@ -13,7 +13,7 @@ def compute_normalized_PSNR(orignal, data):
 
 class SaveExampleResultImageCallBack(tf.keras.callbacks.Callback):
 
-	def __init__(self, dir_path, train_data_subset, color_space: str, nth_batch_sample: int = 512, grid_size: int = 6,
+	def __init__(self, dir_path, train_data_subset, color_space: str, nth_batch_sample: int = 512, grid_size: int = 6, fileprefix: str = "SuperResolution",
 				 **kwargs):
 		super(tf.keras.callbacks.Callback, self).__init__(**kwargs)
 
@@ -26,6 +26,8 @@ def __init__(self, dir_path, train_data_subset, color_space: str, nth_batch_samp
 		self.color_space = color_space
 		self.grid_size = grid_size
 
+		self.fileprefix = fileprefix
+
 		if not os.path.exists(self.dir_path):
 			os.mkdir(self.dir_path)
 
@@ -35,15 +37,15 @@ def on_epoch_begin(self, epoch, logs=None):
 	def on_epoch_end(self, epoch, logs=None):
 		fig = show_expect_predicted_result(model=self.model, image_batch_dataset=self.trainSet,
 										   color_space=self.color_space, nr_col=self.grid_size)
-		fig.savefig(os.path.join(self.dir_path, "SuperResolution{0}.png".format(epoch)))
+		fig.savefig(os.path.join(self.dir_path, "{0}{1}.png".format(self.fileprefix,epoch)))
 		fig.clf()
 		plt.close(fig)
 
 	def on_train_batch_end(self, batch, logs=None):
 		if batch % self.nth_batch_sample == 0:
 			fig = show_expect_predicted_result(model=self.model, image_batch_dataset=self.trainSet,
 											   color_space=self.color_space, nr_col=self.grid_size)
-			fig.savefig(os.path.join(self.dir_path, "SuperResolution_{0}_{1}.png".format(self.current_epoch, batch)))
+			fig.savefig(os.path.join(self.dir_path, "{0}_{1}_{2}.png".format(self.fileprefix, self.current_epoch, batch)))
 			fig.clf()
 			plt.close(fig)