neelriyer

# adapted from: https://github.com/fastai/course-v3/blob/master/nbs/dl1/lesson7-superres.ipynb

t = data.valid_ds[0][1].data
t = torch.stack([t,t])

def gram_matrix(x):
	n,c,h,w = x.size()
	x = x.view(n, c, -1)
	return (x @ x.transpose(1,2))/(c*h*w)

neelriyer

bs,size=64,128
# bs,size=8,480
arch = models.resnet34

src = ImageImageList.from_folder(path_lr).split_by_rand_pct(0.1, seed=42)

tfms = get_transforms(do_flip=True, flip_vert=True, max_zoom = 1.1, max_lighting=0.2, max_rotate = 10)

def get_data(bs,size=None):

neelriyer

data = get_data(bs=64,size=128)

learn = None
gc.collect()
wd = 1e-3
learn = unet_learner(data, arch, wd=wd, loss_func=feat_loss, callback_fns=LossMetrics,blur=True, norm_type=NormType.Weight)

neelriyer

lr = 1e-2

def do_fit(save_name, lrs=slice(lr), pct_start=0.9, cycles = 10):
	learn.fit_one_cycle(cycles, lrs, pct_start=pct_start)
	learn.save(save_name)
	learn.show_results(rows=2, imgsize=7)

do_fit('1a', slice(lr))

neelriyer

data = get_data(bs=1)

learn.data = data
learn.freeze()
gc.collect()

learn.load('2b')

lr = 1e-6

neelriyer

learn = load_learner("/content/drive/My Drive/video_restorer4/")

def run_inference_images(file, dest):

  img = open_image(file)
  p,img_hr,b = learn.predict(img)

  # Image(img_hr).save(dest)
  # plt.figure(figsize=(25,25))
  Image(img_hr).show(figsize=(25,25))

neelriyer

mport PIL
import glob
import os
from tqdm.notebook import tqdm

render_factor = 40

if os.path.exists('imagepaths.txt'):
  os.remove('imagepaths.txt')

neelriyer

import cv2
import numpy as np
import os

clean = 'seinfeld_inference.mp4'
cap = cv2.VideoCapture(clean)
clean_fps = cap.get(cv2.CAP_PROP_FPS)
print(clean_fps)

pathOut = 'video.mp4'

neelriyer

# download grain video
rm -Rf build
YT_GRAIN_OVERLAY="https://www.youtube.com/watch?v=J_MZb7qTenE"
mkdir -p build
youtube-dl "$YT_GRAIN_OVERLAY" -f mp4 --output "build/grain.mp4"
# invert colors
ffmpeg -loglevel quiet -y -i "build/grain.mp4" -vf negate 'color_inverted.mp4'
# overlay video
ffmpeg \
    -y \

neelriyer

class EmbeddingDot(nn.Module):
	def __init__(self):
		super().__init__()
		self.u.weight.data.uniform_(0,0.05)
		self.m.weight.data.uniform_(0,0.05)