Image segmentation with a U-Net-like architecture - Keras Code Examples

Показать описание

This video will show you how to use a U-Net style ConvNet to map from a 160x160xRGB image of a PET into the same 160x160 dimensional annotation map of each pixel in the image. This involves segmenting the pet from its border and the background.

Thanks for watching! Please subscribe and check out the rest of the Keras Code Examples playlist!

Connor Shorten

Рекомендации по теме

Комментарии

Currently working on my bachelor's thesis using a U-Net. What a coincidence that you post this now... how convenient :D

bogenriederlukas

Thanks for doing these walkthroughs, very helpful!

jrtapley

2:38 it's ordering character by character, not treating 100 as one number but as 1-0-0. That's why 100 comes before 2.

tamerius

Thanks for the video.
There are a few technical points I didn't quite understand. What's the advantage of using the Keras.util.Sequence over tf.data? Is it just the flexibility of doing the loading on your own? From what I've read tf.data outperforms the Sequence approach quite hard. Do you have any preferences for data loading and does it even matter to worry about the data loading specifics in an environment like this one?

Also, what do you think about the use of the Conv2DTranspose layers? As far as I understand, they're usually used for learnable upscaling, whereas here they don't change the layer dimensions.
Are they just a different way of messing with the data to learn more patterns?

One more catch - I've noticed there are two consecutive relu activation layers in the model. Is that a common thing to see?

petrmiculek

how can I get output like classification? I mean if the U_Net can segment it or not.

arfi_

Error: ( Unexpected result of `train_function` (Empty logs). Please use `Model.compile(..., run_eagerly=True)`, or for more information of where went wrong, or file a issue/bug to `tf.keras`) How do we solve this?

JoelPrabhod

So its require a mask image for label? What if with new image input?

jamilal-idrus

Hi sir, I need ur help in this implementation. Am getting this error ( Unexpected result of `train_function` (Empty logs). Please use `Model.compile(..., run_eagerly=True)`, or for more information of where went wrong, or file a issue/bug to `tf.keras`) How to debug

priyaakr

Image segmentation with a U-Net-like architecture - Keras Code Examples

The U-Net (actually) explained in 10 minutes

Image segmentation with a U-Net-like architecture - Keras Code Examples

What is U-NET?

PyTorch Image Segmentation Tutorial with U-NET: everything from scratch baby

UNet for Image Segmentation - What You Need To Know! - Computer Vision

1. Transform Your Image Segmentation with U-NET

Image segmentation with a U Net like architecture

U-Net for Image Segmentation - Research Paper Summary

73 - Image Segmentation using U-Net - Part1 (What is U-net?)

Deep Learning Approaches to Image Segmentation

Convolutional Networks for Biomedical Image Segmentation (U-Net) pt-2

Image Segmentation Tutorial | UNet | Oxford Pet Data | Keras Tensorflow

Image Segmentation with U-net and Keras!

Medical image - Retina Vessel segmentation using Optimized Deep learning U-Net CNN || Python project

Tutorial for Debugging U-net code, Complete Retinal Image Example

Semantic Segmentation of Railway Infrastructure - U-Net Demonstration

U-Net clearly explained | Image Segmentation with AI

Master Image Clean Up with U-NET: Object Segmentation Tutorial

Lec 42: U-Net: Convolutional Networks for Image Segmentation

Image segmentation and U-net

Image segmentation with TensorFlow | U-net model

228 - Semantic segmentation of aerial (satellite) imagery using U-net

Leaf Segmentation Challenge Using UNET

Deep learning-based segmentation for biologists - Combining results from U-Net and Mask R-CNN