import torch.utils.data as data
import torch
import numpy as np
import os
from os import listdir
from os.path import join
from PIL import Image, ImageOps
import random
import pyflow
from skimage import img_as_float
from skimage import color
from random import randrange
import os.path
import cv2
max_flow = 150.0
def is_image_file(filename):
return any(filename.endswith(extension) for extension in [".png", ".jpg", ".jpeg"])
def load_img(filepath, scale):
list=os.listdir(filepath)
list.sort()
rate = 1
#for vimeo90k-setuplet (multiple temporal scale)
#if random.random() < 0.5:
# rate = 2
index = randrange(0, len(list)-(2*rate))
target = [modcrop(Image.open(filepath+'/'+list[i]).convert('RGB'), scale) for i in range(index, index+3*rate, rate)]
h,w = target[0].size
h_in,w_in = int(h//scale), int(w//scale)
target_l = target[1].resize((h_in,w_in), Image.BICUBIC)
input = [target[j].resize((h_in,w_in), Image.BICUBIC) for j in [0,2]]
return input, target, target_l, list
def load_img_test(filepath, scale):
list=os.listdir(filepath)
list.sort()
target = [modcrop(Image.open(filepath+'/'+list[i]).convert('RGB'), scale) for i in range(len(list))]
h,w = target[0].size
h_in,w_in = int(h//scale), int(w//scale)
input = [target[j].resize((h_in,w_in), Image.BICUBIC) for j in [0,len(list)-1]]
return input, list
def load_img_nodown(filepath):
list=os.listdir(filepath)
list.sort()
input = [Image.open(filepath+'/'+list[i]).convert('RGB') for i in [0,len(list)-1]]
return input, list
def get_flow(im1, im2):
im1 = np.array(im1)
im2 = np.array(im2)
im1 = im1.astype(float) / 255.
im2 = im2.astype(float) / 255.
# Flow Options:
alpha = 0.012
ratio = 0.75 #0.95 #0.75
minWidth = 20 #50 #20
nOuterFPIterations = 7
nInnerFPIterations = 1
nSORIterations = 30
colType = 0 # 0 or default:RGB, 1:GRAY (but pass gray image with shape (h,w,1))
u, v, im2W = pyflow.coarse2fine_flow(im1, im2, alpha, ratio, minWidth, nOuterFPIterations, nInnerFPIterations,nSORIterations, colType)
flow = np.concatenate((u[..., None], v[..., None]), axis=2)
#Rescale
flow = rescale_flow(flow,-1,1)
return flow
def rescale_flow(x,max_range,min_range):
#remove noise
x[x > max_flow] = max_flow
x[x < -max_flow] = -max_flow
max_val = max_flow
min_val = -max_flow
return (max_range-min_range)/(max_val-min_val)*(x-max_val)+max_range
def modcrop(img, modulo):
(ih, iw) = img.size
ih = ih - (ih%modulo);
iw = iw - (iw%modulo);
img = img.crop((0, 0, ih, iw))
return img
def get_patch(img_in, img_tar, img_tar_l, patch_size, scale, ix=-1, iy=-1):
(ih, iw) = img_in[0].size
(th, tw) = (scale * ih, scale * iw)
patch_mult = scale #if len(scale) > 1 else 1
tp = patch_mult * patch_size
ip = tp // scale
if ix == -1:
ix = random.randrange(0, iw - ip + 1)
if iy == -1:
iy = random.randrange(0, ih - ip + 1)
(tx, ty) = (scale * ix, scale * iy)
img_in = [j.crop((iy,ix,iy + ip, ix + ip)) for j in img_in]
img_tar = [j.crop((ty,tx,ty + tp, tx + tp)) for j in img_tar]
img_tar_l = img_tar_l.crop((iy,ix,iy + ip, ix + ip))
info_patch = {
'ix': ix, 'iy': iy, 'ip': ip, 'tx': tx, 'ty': ty, 'tp': tp}
return img_in, img_tar, img_tar_l, info_patch
def augment(img_in, img_tar, img_tar_l, flip_h=True, rot=True):
info_aug = {'flip_h': False, 'flip_v': False, 'trans': False}
if random.random() < 0.5 and flip_h:
img_in = [ImageOps.flip(j) for j in img_in]
img_tar = [ImageOps.flip(j) for j in img_tar]
img_tar_l = ImageOps.flip(img_tar_l)
info_aug['flip_h'] = True
if rot:
if random.random() < 0.5:
img_in = [ImageOps.mirror(j) for j in img_in]
img_tar = [ImageOps.mirror(j) for j in img_tar]
img_tar_l = ImageOps.mirror(img_tar_l)
info_aug['flip_v'] = True
if random.random() < 0.5:
img_in = [j.rotate(180) for j in img_in]
img_tar = [j.rotate(180) for j in img_tar]
img_tar_l = img_tar_l.rotate(180)
info_aug['trans'] = True
return img_in, img_tar, img_tar_l, info_aug
class DatasetFromFolder(data.Dataset):
def __init__(self, image_dir, upscale_factor, data_augmentation, file_list, patch_size, transform=None):
super(DatasetFromFolder, self).__init__()
alist = [line.rstrip() for line in open(join(image_dir,file_list))]
self.image_filenames = [join(image_dir,x) for x in alist]
self.upscale_factor = upscale_factor
self.transform = transform
self.data_augmentation = data_augmentation
self.patch_size = patch_size
def __getitem__(self, index):
input, target, target_l, file_list = load_img(self.image_filenames[index], self.upscale_factor)
if self.patch_size != 0:
input, target, target_l, _ = get_patch(input,target,target_l,self.patch_size, self.upscale_factor)
if self.data_augmentation:
input, target, target_l, _ = augment(input, target, target_l)
flow_f = get_flow(input[0],input[1])
flow_b = get_flow(input[1],input[0])
if self.transform:
input = [self.transform(j) for j in input]
target = [self.transform(j) for j in target]
target_l = self.transform(target_l)
flow_f = torch.from_numpy(flow_f.transpose(2,0,1))
flow_b = torch.from_numpy(flow_b.transpose(2,0,1))
return input, target, target_l, flow_f, flow_b, file_list, self.image_filenames[index]
def __len__(self):
return len(self.image_filenames)
class DatasetFromFolderFlow(data.Dataset):
def __init__(self, image_dir, upscale_factor, data_augmentation, file_list, patch_size, transform=None):
super(DatasetFromFolderFlow, self).__init__()
alist = [line.rstrip() for line in open(join(image_dir,file_list))]
self.image_filenames = [join(image_dir,x) for x in alist]
self.upscale_factor = upscale_factor
self.transform = transform
self.data_augmentation = data_augmentation
self.patch_size = patch_size
def __getitem__(self, index):
input, target, target_l, file_list = load_img(self.image_filenames[index], self.upscale_factor)
if self.patch_size != 0:
input, target, target_l, _ = get_patch(input,target,target_l,self.patch_size, self.upscale_factor)
if self.data_augmentation:
input, target, target_l, _ = augment(input, target, target_l)
flow_f = get_flow(input[0],input[1])
flow_b = get_flow(input[1],input[0])
gt_flow_f = get_flow(input[0],target_l) + get_flow(target_l,input[1])
gt_flow_b = get_flow(input[1],target_l) + get_flow(target_l,input[0])
if self.transform:
input = [self.transform(j) for j in input]
target = [self.transform(j) for j in target]
target_l = self.transform(target_l)
flow_f = torch.from_numpy(flow_f.transpose(2,0,1))
flow_b = torch.from_numpy(flow_b.transpose(2,0,1))
gt_flow_f = torch.from_numpy(gt_flow_f.transpose(2,0,1))
gt_flow_b = torch.from_numpy(gt_flow_b.transpose(2,0,1))
return input, target, target_l, flow_f, flow_b, gt_flow_f, gt_flow_b,file_list, self.image_filenames[index]
def __len__(self):
return len(self.image_filenames)
class DatasetFromFolderFlowLR(data.Dataset):
def __init__(self, image_dir, upscale_factor, data_augmentation, file_list, patch_size, transform=None):
super(DatasetFromFolderFlowLR, self).__init__()
alist = [line.rstrip() for line in open(join(image_dir,file_list))]
self.image_filenames = [join(image_dir,x) for x in alist]
self.upscale_factor = upscale_factor
self.transform = transform
self.data_augmentation = data_augmentation
self.patch_size = patch_size
def __getitem__(self, index):
input, target, target_l, file_list = load_img(self.image_filenames[index], self.upscale_factor)
if self.patch_size != 0:
input, target, target_l, _ = get_patch(input,target,target_l,self.patch_size, self.upscale_factor)
if self.data_augmentation:
input, target, target_l, _ = augment(input, target, target_l)
flow_f = get_flow(target[0],target[2])
flow_b = get_flow(target[2],target[0])
gt_flow_f = get_flow(target[0],target[1]) + get_flow(target[1],target[2])
gt_flow_b = get_flow(target[2],target[1]) + get_flow(target[1],target[0])
if self.transform:
target = [self.transform(j) for j in target]
flow_f = torch.from_numpy(flow_f.transpose(2,0,1))
flow_b = torch.from_numpy(flow_b.transpose(2,0,1))
gt_flow_f = torch.from_numpy(gt_flow_f.transpose(2,0,1))
gt_flow_b = torch.from_numpy(gt_flow_b.transpose(2,0,1))
return target, flow_f, flow_b, gt_flow_f, gt_flow_b, file_list, self.image_filenames[index]
def __len__(self):
return len(self.image_filenames)
class DatasetFromFolderTest(data.Dataset):
def __init__(self, image_dir, upscale_factor, file_list, transform=None):
super(DatasetFromFolderTest, self).__init__()
alist = [line.rstrip() for line in open(join(image_dir,file_list))]
self.image_filenames = [join(image_dir,x) for x in alist]
self.upscale_factor = upscale_factor
self.transform = transform
def __getitem__(self, index):
input, file_list = load_img_test(self.image_filenames[index], self.upscale_factor)
flow_f = get_flow(input[0],input[1])
flow_b = get_flow(input[1],input[0])
if self.transform:
input = [self.transform(j) for j in input]
flow_f = torch.from_numpy(flow_f.transpose(2,0,1))
flow_b = torch.from_numpy(flow_b.transpose(2,0,1))
return input, flow_f, flow_b, file_list, self.image_filenames[index]
def __len__(self):
return len(self.image_filenames)
class DatasetFromFolderInterp(data.Dataset):
def __init__(self, image_dir, file_list, transform=None):
super(DatasetFromFolderInterp, self).__init__()
alist = [line.rstrip() for line in open(join(image_dir,file_list))]
self.image_filenames = [join(image_dir,x) for x in alist]
self.transform = transform
def __getitem__(self, index):
input, file_list = load_img_nodown(self.image_filenames[index])
flow_f = get_flow(input[0],input[1])
flow_b = get_flow(input[1],input[0])
if self.transform:
input = [self.transform(j) for j in input]
flow_f = torch.from_numpy(flow_f.transpose(2,0,1))
flow_b = torch.from_numpy(flow_b.transpose(2,0,1))
return input, flow_f, flow_b, file_list, self.image_filenames[index]
def __len__(self):
return len(self.image_filenames)
