Deep Image Enhancement Project Progress Report

undefined
 
DEEP SCREEN IMAGE CROP
AND ENHANCE
 
Week 2
(Aaron Ott, Amir Mazaheri)
 
Problem
 
We have taken a photo of an image,
and we want the original image.
 
This can be broken into 2 parts:
Image Detector/Cropper
Image Enhancer
 
Enhancer
 
https://github.com/krasserm/super-resolution
 
 - Pretrained EDSR (trained on DIV2K)
 - Modified form of Resnet
 - Actually scales up the image 4x
 
Lim, Son, Kim, Nah, Lee. “Enhanced Deep Residual Networks for Single Image Super-Resolution”. 10 July 2017
 
Combining the Cropper and Enhancer
 
Had to use lambda functions and
AveragePooling2D to get the Enhancer to
properly work with the Cropper
 
Loss Functions:
 
-
 
Trained Cropper on VGG+CosineProximity
 
-
 
Trained Enhancer on VGG+MSE
 
Last Week
(Cropper)
 
Cropper using a Spatial Transformation Network
 
Input
 
Output
 
Actual
Metrics:
 
PSNR: 
 
11.1903
 
SSIM:
 
0.4254
 
MSE:
 
0.0796
 
This Week
(Cropper + Enhancer)
Metrics:
 
PSNR: 
 
15.8368
 
SSIM:
 
0.4807
 
MSE:
 
0.0316
 
Input
 
Output
 
Actual
 
Metrics Compared
 
Shortcomings of PSNR, SSIM, and MSE
 
Shortcomings of PSNR, SSIM, and MSE
Metrics:
 
PSNR: 
 
18.2626
 
SSIM:
 
0.5314
 
MSE:
 
0.0166
 
Input
 
Output
 
Actual
 
Loss Functions
VGG + Cosine Proximity
MSE
 
Next Week
 
Create and refine GAN architecture
Refine EDSR or try a different SISR (WDSR)
Slide Note
Embed
Share

The Deep Screen Image Crop and Enhance project, led by Aaron Ott and Amir Mazaheri, focuses on improving image quality through a multi-step approach involving image detection, cropping, and enhancement. The project utilizes advanced techniques like super-resolution networks and deep residual networks to enhance image quality. Progress metrics show significant improvements in image quality metrics like PSNR and SSIM. Challenges and future plans include refining GAN architecture and exploring different single image super-resolution methods.


Uploaded on Aug 01, 2024 | 0 Views


Download Presentation

Please find below an Image/Link to download the presentation.

The content on the website is provided AS IS for your information and personal use only. It may not be sold, licensed, or shared on other websites without obtaining consent from the author. Download presentation by click this link. If you encounter any issues during the download, it is possible that the publisher has removed the file from their server.

E N D

Presentation Transcript


  1. DEEP SCREEN IMAGE CROP AND ENHANCE Week 2 (Aaron Ott, Amir Mazaheri)

  2. Problem We have taken a photo of an image, and we want the original image. This can be broken into 2 parts: Image Detector/Cropper Image Enhancer

  3. Enhancer https://github.com/krasserm/super-resolution - Pretrained EDSR (trained on DIV2K) - Modified form of Resnet - Actually scales up the image 4x Lim, Son, Kim, Nah, Lee. Enhanced Deep Residual Networks for Single Image Super-Resolution . 10 July 2017

  4. Combining the Cropper and Enhancer Had to use lambda functions and AveragePooling2D to get the Enhancer to properly work with the Cropper Loss Functions: - - Trained Cropper on VGG+CosineProximity Trained Enhancer on VGG+MSE

  5. Metrics: Metrics: Last Week (Cropper) PSNR: 11.1903 Cropper using a Spatial Transformation Network SSIM: 0.4254 MSE: 0.0796 Input Output Actual

  6. Metrics: Metrics: This Week (Cropper + Enhancer) PSNR: 15.8368 SSIM: 0.4807 MSE: 0.0316 Input Output Actual

  7. Metrics Compared Model Model \ \ Metric Metric PSNR PSNR SSIM SSIM MSE MSE Cropper 11.1903 0.4254 0.0796 Cropper + Enhancer 15.8368 15.8368 0.4807 0.4807 0.0316 0.0316

  8. Shortcomings of PSNR, SSIM, and MSE Model Model \ \ Metric Metric PSNR PSNR SSIM SSIM MSE MSE Cropper 11.1903 0.4254 0.0796 Cropper + Enhancer 15.8368 0.4807 0.0316 Cropper + Enhancer w/ MSE Loss 18.2626 18.2626 0.5314 0.5314 0.0166 0.0166

  9. Shortcomings of PSNR, SSIM, and MSE Metrics: Metrics: PSNR: 18.2626 Input SSIM: 0.5314 MSE: 0.0166 Output Loss Functions VGG + Cosine Proximity MSE Actual

  10. Next Week Create and refine GAN architecture Refine EDSR or try a different SISR (WDSR)

Related


More Related Content

giItT1WQy@!-/#giItT1WQy@!-/#giItT1WQy@!-/#