Mask RCNN Tutorial Training Pothole Segmentation | OpenCV Python | Computer Vision 2020 Part 3

preview_player
Показать описание
Mask RCNN Tutorial #3 - Training Mask RCNN for Pothole Segmentation - Dataset & Annotation - We use OpenCV and Python for our final implementation.

Hey guys and welcome back, in the next few videos, I’m going to show you how to build your own pot-hole detector using Mask RCNN. Now Mask RCNN is particularly useful for this application because you are able to quantify the amount of pixels belong the pothole thus allowing you to measure how big and how bad the pothole is. In most countries potholes and sink holes are quite a huge problem and costs each countries more than a billion dollars a year to fix.

Now imagine if you had a scout vehicle or drone that is able to automatically geolocate, these potholes. They would be able to measure them using image data to then determine if the road needs to be fixed or even better to predict when and where the next pot hole will occur.

So in this lecture, we are going to start with downloading the dataset. You also have the option to scrape your pot hole image dataset from google image search just like how we did in the Yolo V3 course. We perform mask rcnn pytorch tutorial in this lecture.

--------------------------------------------------------------
Geeky Bee AI Pvt Ltd
------------------------------------------------------------
Support Augmented Startups on Patreon
Chat to us on Discord
Interact with us on Facebook
Check my latest work on Instagram
Learn Advanced Tutorials on Udemy
------------------------------------------------------------
To learn more on Artificial Intelligence, Augmented Reality IoT, Deep Learning FPGAs, Arduinos, PCB Design and Image Processing then check out

Please Like and Subscribe for more videos :)
Рекомендации по теме
Комментарии
Автор

Can the model learn to segment or detect potholes of various sizes? Or only if your training set included a pothole of every possible size?

Robotwhisperer
Автор

May I know how to train a model without using any pre-trained weights like from coco or imagenet I wish to train the model from scratch with no transfer learning how can I do it? It will be great help thanks!

akshayijantkar
Автор

You should inform people that if they use your annotation tool they will have very hard time in converting it in any other format so they can train their models locally or on google colab not only on supervisely. I have lost a lot of time with that.

kiramarc
Автор

Other than Mask RCNN what all other models can be used for Instance Object Segmentation? Thanks in advance!

akshayijantkar
Автор

I get this error when i tried to training my dataset UserWarning: Using a generator with `use_multiprocessing=True` and multiple workers may duplicate your data. Please consider using the`keras.utils.Sequence class.

abdechafineji
Автор

What is the reason for START IMPORT button disable ?
Thank you!

mihirichathurikaamarasingh
Автор

Being distracted, I heard « in the next video, I’ll show you how to detect buttholes. » at first. I burst out laughing.... :D

mahrard
Автор

supervisely says link not found for the annotated dataset

shubhamgupto
Автор

AI phân biệt lưu trữ nhóm và thói quen hành vi

HungPham-pgoq
Автор

hello sir,
Your all videos are amazing, I learnt a lot from every single video, now
I am able to make any AR app within a minutes.
I have a question about Video stability and it's framing. how to set
framing in any video? and how can i manage the stabilility of any
videos on visiting card ?

brijeshkapadiya