More cleaning

codes/data/__init__.py

@@ -33,33 +33,33 @@ def create_dataset(dataset_opt, return_collate=False):
 
     # datasets for image restoration
     if mode == 'fullimage':
-        from data.full_image_dataset import FullImageDataset as D
+        from data.images.full_image_dataset import FullImageDataset as D
     elif mode == 'single_image_extensible':
-        from data.single_image_dataset import SingleImageDataset as D
+        from data.images.single_image_dataset import SingleImageDataset as D
     elif mode == 'multi_frame_extensible':
-        from data.multi_frame_dataset import MultiFrameDataset as D
+        from data.images.multi_frame_dataset import MultiFrameDataset as D
     elif mode == 'combined':
         from data.combined_dataset import CombinedDataset as D
     elif mode == 'multiscale':
-        from data.multiscale_dataset import MultiScaleDataset as D
+        from data.images.multiscale_dataset import MultiScaleDataset as D
     elif mode == 'paired_frame':
-        from data.paired_frame_dataset import PairedFrameDataset as D
+        from data.images.paired_frame_dataset import PairedFrameDataset as D
     elif mode == 'stylegan2':
-        from data.stylegan2_dataset import Stylegan2Dataset as D
+        from data.images.stylegan2_dataset import Stylegan2Dataset as D
     elif mode == 'imagefolder':
-        from data.image_folder_dataset import ImageFolderDataset as D
+        from data.images.image_folder_dataset import ImageFolderDataset as D
     elif mode == 'torch_dataset':
         from data.torch_dataset import TorchDataset as D
     elif mode == 'byol_dataset':
-        from data.byol_attachment import ByolDatasetWrapper as D
+        from data.images.byol_attachment import ByolDatasetWrapper as D
     elif mode == 'byol_structured_dataset':
-        from data.byol_attachment import StructuredCropDatasetWrapper as D
+        from data.images.byol_attachment import StructuredCropDatasetWrapper as D
     elif mode == 'random_aug_wrapper':
-        from data.byol_attachment import DatasetRandomAugWrapper as D
+        from data.images.byol_attachment import DatasetRandomAugWrapper as D
     elif mode == 'random_dataset':
-        from data.random_dataset import RandomDataset as D
+        from data.images.random_dataset import RandomDataset as D
     elif mode == 'zipfile':
-        from data.zip_file_dataset import ZipFileDataset as D
+        from data.images.zip_file_dataset import ZipFileDataset as D
     elif mode == 'nv_tacotron':
         from data.audio.nv_tacotron_dataset import TextWavLoader as D
         from data.audio.nv_tacotron_dataset import TextMelCollate as C

codes/data/images/base_unsupervised_image_dataset.py

@@ -1,7 +1,7 @@
 import torch
 from torch.utils import data
-from data.image_corruptor import ImageCorruptor
-from data.chunk_with_reference import ChunkWithReference
+from data.images.image_corruptor import ImageCorruptor
+from data.images.chunk_with_reference import ChunkWithReference
 import os
 import cv2
 import numpy as np

codes/data/images/image_folder_dataset.py

@@ -1,24 +1,19 @@
 import functools
-import glob
-import itertools
 import random
 
 import cv2
-import kornia
 import numpy as np
-import pytorch_ssim
 import torch
-import os
 
 import torchvision
 from torch.utils.data import DataLoader
-from torchvision.transforms import Normalize, CenterCrop
+from torchvision.transforms import Normalize
 from tqdm import tqdm
 
 from data import util
 # Builds a dataset created from a simple folder containing a list of training/test/validation images.
-from data.image_corruptor import ImageCorruptor, kornia_color_jitter_numpy
-from data.image_label_parser import VsNetImageLabeler
+from data.images.image_corruptor import ImageCorruptor, kornia_color_jitter_numpy
+from data.images.image_label_parser import VsNetImageLabeler
 from utils.util import opt_get
 
 

codes/data/images/image_pair_with_corresponding_points_dataset.py

@@ -1,11 +1,3 @@
-import glob
-import itertools
-import random
-
-import cv2
-import kornia
-import numpy as np
-import pytorch_ssim
 import torch
 import os
 
@@ -13,14 +5,10 @@ import torchvision
 from PIL import Image
 from torch.utils.data import DataLoader, Dataset
 from torchvision import transforms
-from torchvision.transforms import Normalize
 from tqdm import tqdm
 
-from data import util
+
 # Builds a dataset created from a simple folder containing a list of training/test/validation images.
-from data.image_corruptor import ImageCorruptor
-from data.image_label_parser import VsNetImageLabeler
-from utils.util import opt_get
 
 
 class ImagePairWithCorrespondingPointsDataset(Dataset):
@@ -70,7 +58,7 @@ if __name__ == '__main__':
         'path': 'F:\\dlas\\codes\\scripts\\ui\\image_pair_labeler\\results',
         'size': 256
     }
-    output_path = '.'
+    output_path = '..'
 
     ds = DataLoader(ImagePairWithCorrespondingPointsDataset(opt), shuffle=True, num_workers=0)
     for i, d in tqdm(enumerate(ds)):

codes/data/images/multi_frame_dataset.py

@@ -1,4 +1,4 @@
-from data.base_unsupervised_image_dataset import BaseUnsupervisedImageDataset
+from data.images.base_unsupervised_image_dataset import BaseUnsupervisedImageDataset
 import numpy as np
 import torch
 from bisect import bisect_left

codes/data/images/multiscale_dataset.py

@@ -4,13 +4,9 @@ import cv2
 import torch
 import torch.utils.data as data
 import data.util as util
-from PIL import Image, ImageOps
-from io import BytesIO
-import torchvision.transforms.functional as F
-
 
 # Reads full-quality images and pulls tiles at regular zoom intervals from them. Only usable for training purposes.
-from data.image_corruptor import ImageCorruptor
+from data.images.image_corruptor import ImageCorruptor
 
 
 # Selects the smallest dimension from the image and crops it randomly so the other dimension matches. The cropping

codes/data/images/paired_frame_dataset.py

@@ -1,4 +1,4 @@
-from data.base_unsupervised_image_dataset import BaseUnsupervisedImageDataset
+from data.images.base_unsupervised_image_dataset import BaseUnsupervisedImageDataset
 import numpy as np
 import torch
 from bisect import bisect_left

codes/data/images/single_image_dataset.py

@@ -2,8 +2,7 @@ import random
 from bisect import bisect_left
 import numpy as np
 import torch
-from torch.utils import data
-from data.base_unsupervised_image_dataset import BaseUnsupervisedImageDataset
+from data.images.base_unsupervised_image_dataset import BaseUnsupervisedImageDataset
 
 
 # Builds a dataset composed of a set of folders. Each folder represents a single high resolution image that has been

codes/data/torch_dataset.py

@@ -1,10 +1,9 @@
-import torch
 from torch.utils.data import Dataset
 import torchvision.transforms as T
 from torchvision import datasets
 
 # Wrapper for basic pytorch datasets which re-wraps them into a format usable by ExtensibleTrainer.
-from data.cifar import CIFAR100, CIFAR10
+from data.images.cifar import CIFAR100, CIFAR10
 from utils.util import opt_get
 
 

codes/models/image_latents/byol/byol_model_wrapper.py

@@ -9,7 +9,7 @@ import torchvision
 from kornia import filters
 from torch import nn
 
-from data.byol_attachment import RandomApply
+from data.images.byol_attachment import RandomApply
 from trainer.networks import register_model, create_model
 from utils.util import checkpoint, opt_get
 

codes/models/image_latents/byol/byol_structural.py

@@ -4,7 +4,7 @@ import torch
 import torch.nn.functional as F
 from torch import nn
 
-from data.byol_attachment import reconstructed_shared_regions
+from data.images.byol_attachment import reconstructed_shared_regions
 from models.image_latents.byol.byol_model_wrapper import singleton, EMA, get_module_device, set_requires_grad, \
     update_moving_average
 from trainer.networks import create_model, register_model

codes/scripts/byol/byol_resnet_playground.py

@@ -9,7 +9,7 @@ from torch.utils.data import DataLoader
 from torchvision.transforms import ToTensor
 from tqdm import tqdm
 
-from data.image_folder_dataset import ImageFolderDataset
+from data.images.image_folder_dataset import ImageFolderDataset
 from models.classifiers.resnet_with_checkpointing import resnet50
 
 # Computes the structural euclidean distance between [x,y]. "Structural" here means the [h,w] dimensions are preserved

codes/scripts/byol/byol_segformer_playground.py

@@ -9,7 +9,7 @@ from torch.utils.data import DataLoader
 from torchvision.transforms import ToTensor, Normalize
 from tqdm import tqdm
 
-from data.image_folder_dataset import ImageFolderDataset
+from data.images.image_folder_dataset import ImageFolderDataset
 from models.segformer.segformer import Segformer
 
 # Computes the structural euclidean distance between [x,y]. "Structural" here means the [h,w] dimensions are preserved

codes/scripts/byol/byol_spinenet_playground.py

@@ -10,7 +10,7 @@ from torchvision.transforms import ToTensor, Resize
 from tqdm import tqdm
 import numpy as np
 
-from data.image_folder_dataset import ImageFolderDataset
+from data.images.image_folder_dataset import ImageFolderDataset
 from models.image_latents.spinenet_arch import SpineNet
 
 

codes/scripts/srflow_latent_space_playground.py

@@ -13,7 +13,7 @@ from tqdm import tqdm
 import utils.options as option
 
 import utils
-from data.image_corruptor import ImageCorruptor
+from data.images.image_corruptor import ImageCorruptor
 from trainer.ExtensibleTrainer import ExtensibleTrainer
 from utils import util
 

codes/scripts/ui/image_labeler/image_labeler_ui.py

@@ -11,7 +11,7 @@ import torch
 import torchvision
 from PIL import ImageTk
 
-from data.image_label_parser import VsNetImageLabeler
+from data.images.image_label_parser import VsNetImageLabeler
 from scripts.ui.image_labeler.pretrained_image_patch_classifier import PretrainedImagePatchClassifier
 
 # Globals used to define state that event handlers might operate on.

codes/scripts/ui/image_labeler/label_editor.py

@@ -1,6 +1,6 @@
 import orjson
 
-from data.image_label_parser import VsNetImageLabeler
+from data.images.image_label_parser import VsNetImageLabeler
 
 
 # Translates from the label JSON output of the VS.NET UI to something more compact and usable.

codes/scripts/use_generator_as_filter.py

@@ -3,7 +3,7 @@ import shutil
 
 from torch.utils.data import DataLoader
 
-from data.single_image_dataset import SingleImageDataset
+from data.images.single_image_dataset import SingleImageDataset
 from tqdm import tqdm
 import torch
 

codes/trainer/custom_training_components/progressive_zoom.py

@@ -5,7 +5,7 @@ import torch
 import torchvision
 from torch.cuda.amp import autocast
 
-from data.multiscale_dataset import build_multiscale_patch_index_map
+from data.images.multiscale_dataset import build_multiscale_patch_index_map
 from trainer.inject import Injector
 from trainer.losses import extract_params_from_state
 import os.path as osp

codes/trainer/eval/flow_gaussian_nll.py

@@ -5,7 +5,7 @@ from tqdm import tqdm
 import trainer.eval.evaluator as evaluator
 
 # Evaluate how close to true Gaussian a flow network predicts in a "normal" pass given a LQ/HQ image pair.
-from data.image_folder_dataset import ImageFolderDataset
+from data.images.image_folder_dataset import ImageFolderDataset
 from models.image_generation.srflow import GaussianDiag
 
 

codes/trainer/eval/single_point_pair_contrastive_eval.py

@@ -1,18 +1,13 @@
-import os
-
 import torch
-import os.path as osp
-import torchvision
 from torch.nn import MSELoss
 from torch.utils.data import DataLoader
 from tqdm import tqdm
 
 import trainer.eval.evaluator as evaluator
-from pytorch_fid import fid_score
 
-from data.image_pair_with_corresponding_points_dataset import ImagePairWithCorrespondingPointsDataset
+from data.images.image_pair_with_corresponding_points_dataset import ImagePairWithCorrespondingPointsDataset
 from models.segformer.segformer import Segformer
-from utils.util import opt_get
+
 
 # Uses two datasets: a "similar" and "dissimilar" dataset, each of which contains pairs of images and similar/dissimilar
 # points in those datasets. Uses the provided network to compute a latent vector for both similar and dissimilar.

codes/trainer/eval/sr_style.py

@@ -11,7 +11,7 @@ from pytorch_fid import fid_score
 
 # Evaluate that feeds a LR structure into the input, then calculates a FID score on the results added to
 # the interpolated LR structure.
-from data.stylegan2_dataset import Stylegan2Dataset
+from data.images.stylegan2_dataset import Stylegan2Dataset
 
 
 class SrStyleTransferEvaluator(evaluator.Evaluator):