Presentation is loading. Please wait.

Presentation is loading. Please wait.

Image Inpainting Using Pre-Trained Classification CNN

Similar presentations


Presentation on theme: "Image Inpainting Using Pre-Trained Classification CNN"— Presentation transcript:

1 Image Inpainting Using Pre-Trained Classification CNN
By - Yaniv Kerzhner & Adar Elad Supervisor - Dr. Yaniv Romano

2 The process of reconstructing lost parts in a given image
The Problem - Image Inpainting The process of reconstructing lost parts in a given image M This project aims to apply convolutional neural networks (CNN) for solving the inpainting problem Our prime goal - Investigate whether a pre-trained CNN classifier (for digits/faces) has memorized the data in order to fill it in Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

3 CNN – Convolutional Neural Networks
Input Output Lena Bengio Abbel Hinton Le-Cun Lin 1% 90% 2% Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

4 The straight-forward solution
Given a set of training pairs (original and corrupted images), train a network in a supervised fashion Neural Network , Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

5 The straight-forward solution
Given a set of training pairs (original and corrupted images), train a network in a supervised fashion Nian Cai. et. al. (2015) J. Xie et. al. (2012) D. Pathak et. al. (2016) Raymond A, et. al. (2017) Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

6 The Proposed Method Pre-trained CNN car house lizard tree
Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

7 The Proposed Method Pre-trained CNN car house lizard tree
Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

8 The Proposed Method Pre-trained CNN car house lizard tree
Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

9 The Proposed Method Pre-trained CNN car house lizard tree
Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

10 The Proposed Method Pre-trained CNN car house lizard tree
Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

11 The Proposed Method The Proposed Method Adar Elad & Yaniv Kerzhner
The Computer-Science Department The Technion

12 Results – MNIST dataset
original Masked Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Our Result Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

13 Results – Yale Extended B dataset
original Masked Our Result Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

14 90% of the image is missing !!
Results – ImageNet dataset 90% of the image is missing !! Masked Our Result Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source. Adar Elad & Yaniv Kerzhner The Computer-Science Department The Technion

15 Results – Special Cases
Hallucination - The specific case of image inpainting where all the data is missing 3 5 8 Illumination - Many face images in the dataset were taken in the dark, the question then arises: could these images be illuminated with the assistant of our algorithm? input It is possible to distinguish that all the details of the original image still remains after the inpainting. Meaning, all we did is to add information to the dark parts. our result Suggested Interaction Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source.

16 Time to Conclude Time to Conclude Suggested Interaction
Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source.

17 Time to Conclude Suggested Interaction
Timing: After showing the entire content of this slide Question: Will this algorithm operate well on radar (SAR) images (show such images)? Problem Type: Checkbox Problem with answers YES or NO, and explain Problem Type: Checkbox Problem with answers YES or NO, and explain in a discussion Then show the answer – No, because the model does not match this data source.


Download ppt "Image Inpainting Using Pre-Trained Classification CNN"

Similar presentations


Ads by Google