Single Shot Detector | SSD | Object Detection Using SSD

Explained what is Single Shot Detector.
You can learn other object detection algorithms from below given link:
Yolo Algorithm: • YOLO - Object Detecti...
Practical Implementation of Yolo: • Object Detection Using...
Faster R-CNN: • 1 Object Detection Usi...
Practical Implementation of Faster R-CNN: • 2 Faster R-CNN | Objec...
You can Contact me at : [e-mail]
If you have any questions with what we covered in this video then feel free to ask in the comment section below & I'll do my best to answer your queries.
Please consider clicking the SUBSCRIBE button to be notified for future videos & thank you all for watching.
Channel: www.youtube.co....
Support my channel 🙏 by LIKE ,SHARE & SUBSCRIBE
Check the complete Machine Learning Playlist : www.youtube.co....
Check the complete Deep Learning Playlist : www.youtube.co....
Subscribe my channel: www.youtube.co....
Support my channel 🙏 by LIKE ,SHARE & SUBSCRIBE
Single Shot Detector(SSD) - Real Time Object Detection
Object Detection Is A Technique In Computer Vision That Deals With Detecting Examples Of Semantic Targets Of A Specific Class (Eg. Cars, Buildings Or Humans) In Images And Videos. It Is A Technique That Works To Locate And Identify Objects In Digital Images And Videos.
It Specifically Draws Bounding Boxes Around The Object Which Help Us To Locate Where The Objects Are. Many A Times Object Detection Is Mix With Image Recognition.
SSD
There Are Many Object Detection Algorithms In Practice Like R-CNN, Fast R-CNN, Faster R-CNN Etc..
But Single Shot Algorithms More Efficient And Have A Good Accuracy. They Use Deep Learning Based Approaches For Object Detection.
How Single Shot Detection(SSD) Is Different:-
Single Shot Detection - This Means That The Tasks Of Object Localization And Object Classification Are Ready In A Single Forward Pass Of The Network.
Detector - The Network Is A Detector That Also Classifies The Detected Objects.
Single Shot Detector Is Faster Than The Previous State-Of-The-Art Techniques(YOLO) And Is Significantly More Accurate.
SSD Predicts Category Scores And Box Offsets For A Fixed Number Of Default Bounding Boxes Using Convolution Filters Applied To Feature Maps.
To Achieve High Accuracy We Produce Predictions Of Different Scales From Feature Maps Of Different Dimensions, And Then Separate The Predictions By Aspect Ratio.
These Features Lead To High Accuracy, Even On Low Resolution Input Images.
Other Algorithms Normally Use Object Proposal Methodology Where They Would Come Up With A Way To Break Down The Image Segmented Into Parts To Suggest Where They Could Potentially Be Objects. These Algorithms Sacrifice Accuracy.
Therefore Researchers Came Up With An Interesting Solution Where They Do Everything In One Single Shot. It Just Looks At The Image Once, It Doesn’t Have To Go Back To The Image Again, It Doesn’t Have To Run Many Convolutional Neural Networks.
#SingleShotDetector #SSD #ObjectDetection #PifordTechnologies #AI #ArtificialIntelligence #DeepLearning #ConvolutionalNeuralNetwork #CNN #ComputerVision

[@村田淳七八]

Thank you for finding the explanation of SSD helpful.May I ask you a simple question? After passing through the layer in the 6-layer model, are there no more convolutional layers?

[@MyungeinHan]

Thanks for your video!

[@CodeWithAarohi]

Glad it was helpful!

[@mohamedfawzy2901]

Thanks from Egypt!

[@CodeWithAarohi]

Welcome

[@soravsingla6574]

Best one

[@CodeWithAarohi]

Thanks

[@neenadkambli3168]

great video but i had a different doubt in ssd mobilenet why is mobilenet used if ssd can do both detection and classification of objects in image

[@ramkumarsakthivel9412]

Mobilenet is used to extract the feature maps. SSD's architecture takes in a feature map that is obtained after running either Mobilenet, Resnet, Vggnet, etc on the input image.

[@soravsingla8782]

Very easy to understand.

[@devkhant4505]

Great Video

[@CodeWithAarohi]

Thankyou

[@yontenjamtsho4315]

Thank you for the excellent video. How SDD produces 8732 BB of different sizes? Is the algorithm generating those boxes?

[@CodeWithAarohi]

The number 8732 refers to the total number of default boxes or anchor boxes that SSD generates across the image. These default boxes act as potential bounding box predictions at different scales and aspect ratios.

[@arnavthakur5409]

Thank you mam

[@CodeWithAarohi]

Welcome

[@sohamroykarmokar3071]

This is a great video! Can you make a video on RefineDet: Single-Shot Refinement Neural Network for Object Detection ?

[@CodeWithAarohi]

I will try!

[@awaisahmad5908]

You are best teacher. I have learned a lot from your lectures. I am also research student and my discipline of research is computer vision. Thank you so much.

[@CodeWithAarohi]

Glad my videos are helpful! Keep watching😊

[@yeshwantkumar8181]

Amazing, Thats great. I am the one who is the 100th liker of this

[@CodeWithAarohi]

Thankyou

[@Maria-wm2bb]

I think that ssd makes 8732 prediction for each image not for the class.

[@last_theorem]

so here we are looking for a bounding box with the highest probablity score for the class and the box . And this is done with the anchor box. Isn't it a less efficient way. Because your object can be of different size and shape so you need bounding boxes of different size and shape so in one way are talking about generating several thousands of bounding boxes and checking the intersection of them it will be a brute force approach. And what if there are boxes with no intersection and .

[@CodeWithAarohi]

Yes object can be of different shapes and sizes - to handle this problem we are using bounding boxes of different aspect ratios and sizes so that every size of object get covered. And if there is no intersection that simply means it is a background. And about the approach: we are studying what is SSD and SSD is the algorithm which perform object detection by the explained method. And I have explained what is there in the original paper of SSD.

[@aarondurante1385]

Thank you so much Maam. I am learning a lot. Hope you will create more videos related to this. 👏

[@CodeWithAarohi]

Glad to know...welcome😊

[@sadekasany949]

Thank you ma'am. I can easily understand ur tutarial. So please upload mask rcnn tutarial. Plz ma'am

[@CodeWithAarohi]

Sure

[@elektroprogramming]

First , I want to thank you. You make this video, it helps me a lot. I played multiple times. So, also I have a question You said that after we have extracted feature map , the next process is SSD that have 8732 in every object. but in 4:20 , you showed me that the image predict with boxes is using the real input image(300x300) , not using extracted feauture map . I need an explaination about it. Thanks a lot

[@CodeWithAarohi]

yeah I have used the same image for explanation purpose but you can assume it as a feature map. Sorry that I didn't cleared it there.

[@elektroprogramming]

@@CodeWithAarohi questions again, in the extra feature layers there are 6 layer with different dimension. every single layer do scan an image matching with the ground truth right? is the aspect ratio at layer 1 have the same size (just 1) ? or layer 1 have many size aspect ratio ? so What about next layer2 - layer6 ? what's the relationship between layer1(38*38) and aspect ratio? Many thank's

[@MultiZe10]

Thank you for the explanation, very helpful video. Do you have references or papers that complement the theoric part of SSD? I'm writting a thesis about CNN and need more theorical info about SSD.

[@CodeWithAarohi]

Sorry I don't have an references or papers

[@anamikamaurya22]

sir if you found the references please share

[@ilyasmalik714]

What is confidence code?

[@hussainasghar2841]

can you please send the object detection algorithm for SSD please..

[@CodeWithAarohi]

I will post today.

[@CodeWithAarohi]

kzbin.info/www/bejne/i6jcq3ibn9p6jc0

[@priyalgeorge7112]

the explaination is good, but the sound quality is poor

[@CodeWithAarohi]

Sorry for inconvenience. I have improved the voice quality in my new videos. Will try to redo this video.

[@priyalgeorge7112]

👍

[@hamzanaeem4838]

You have brilliantly explained but , you need to expand more contents that why we are using these 6 layers , why not 5 etc and what is the function of it , How it extract the features whether they are scale or rotation invariant or not etc etc ?

[@CodeWithAarohi]

Thanks for appreciating my work.We are using 6 layers and these 6 layers are prediction layers and it is as per original paper of SSD .You can even use 5 or 4 or 3 prediction layers but here we are using 6 as per the paper of SSD.

[@last_theorem]

I dont have much idea about the VGG, but I have read about the ResNet50, You asked an interesting question. So what exactly is the logic of this choice, what happens in the case of convolutions is that as you go deeper there will be a lot of feature loss, Only the most significant features pop out. So consider a situation when you have tiny objects to detect in a big image. It will be a hard problem since it's not necessary that we can get all the features to detect the image object. We don't choose layers like 1-2 since there are too many features there.( I recommend you to see some visualization of how trained conv looks like.) . The section part is called the feature pyramid since its converging to on point. So we make a prediction from each of the feature layers for the class and the box. The box with the highest confidence and highest class probability is chosen. We choose this pyramid because we get prediction of objects which are small in the image. I have confusion regarding how the bounding boxes work but I think if you put some time into ths solutions are there in the implementation of object detection in Keras . You can read more about it in a paper called feature pyramid , or read the retinanet paper all the details are there. If you want you can try out a different parts of the pyramid, that you can consider your own version, its just a matter of reconfiguring the network. Maybe in some Kaggle competition, you can try that as well.

[@christabelajaero7983]

thank you for this video. it helped alot. Please can you recommend some papers or article i can read to know more about the theoretical part of SSD? 2. How can i get the SDD Convuluntional neural network (Detector network ) which i should use on a pretrained network, say rasnet-50 network.??? I want to try it on matlab. Many Thanks again!!!

[@CodeWithAarohi]

Mail me at [e-mail]. I will send you python code there (SSD with resnet50)

[@anamikamaurya22]

@@CodeWithAarohi Is VGG-16 network is not sufficient.

[@veerababu1033]

voice is very low pls increase it

[@anandtewari8014]

voice is very low in this video...

[@CodeWithAarohi]

Sorry for inconvenience. Took care in my upcoming videos :)

[@mohammedbari3826]

there is no clarity in your explanation. sometimes you are saying 8732 per object and sometime per image.

[@CodeWithAarohi]

Thanks for pointing that out! I'll try to be more clear in future videos.

[@WIngle-jm4qe]

this video is silent af

[@thelaughtermedia9702]

please improve the voice i mean be louder

[@CodeWithAarohi]

Sorry for inconvenience and actually I have improved the voice now but thank you for the feedback :)

[@SourabhSoni-fh4gu]

u need a good mic

[@CodeWithAarohi]

Sorry for the inconvenience and yes I have changed the mic :)