Валентин Сичкар

Пікірлер

@rishabhgupta734 2 жыл бұрын

One question, is ground truth bounding box and anchor boxes used here interchangeably?

@pulkitverma1507 2 жыл бұрын

Very helpful thanks!

@seolakim5667 2 жыл бұрын

Thank you so much for this amazing video. Just one question : at 23:58 , why would you define the "t_0" inside the sigmoid? In the loss function of Yolo v3 they directly use p_0 so I would like to know why! Is this just to make sure that the p_0 is between 0 and 1? Does this t_0 appear somewhere in the model when we implement it? Thanks in advance to anyone who would reply :)

@Can-ue7de 2 жыл бұрын

Amazing Explanation of Yolo v3. Thank you very much.

@sekharbabu8498 2 жыл бұрын

Good explanation. Thank you sir

@bharathnvadla 2 жыл бұрын

Hi Thank you for the explanation ,I have one question, How is the Objectiveness score calculated during the inference ? There is no groundtruth to refer to, on what basis the objectiveness score is measured ?

@tellmebaby183 3 жыл бұрын

this is deep and fantastic, i call for vodka shots

@sameershaik7250 3 жыл бұрын

Explained very well.... great

@Илья96-с7б 3 жыл бұрын

Топчик просто. Сразу всё понятно стало. Стало хоть ясно, что за якоря такие

@ahhhwhysocute 3 жыл бұрын

Amazing explanation !! Thank you

@mainulalam7767 3 жыл бұрын

Thank you for this super explanation. I have a question regarding the objectness score. As you explained mathematically : P0 = sigmoid ( to) = P(object) * IoU -> my question is how we obtain this "P(object)" - predicted probability ? Thanks in advance for your support ..

@bharath5666 3 жыл бұрын

yes,it is predicted probability by the network.

@jessmendoza1483 2 жыл бұрын

@@bharath5666 can i find how does the network predices P(object), but like mathematically or somewhere in the code?

@kyawnaingwin8300 3 жыл бұрын

Should the input image for detection be same size as training images used in model fitting? Or how big is an input image size ok?

@valentynsichkar 3 жыл бұрын

Hello there, There is no need to resize images before training or testing after training. The framework (e.g. the one on GitHub framework for YOLO) will take care of resizing. Moreover, separate images, both for training and testing, can be also of different dimensions.

@kyawnaingwin8300 3 жыл бұрын

@@valentynsichkar thanks for reply. In my case my test image is 20,000 x 20,000 size (drone photo mosaic) and model cannot detect. Only when I split the input image as tiles of same size of training images, it work. According to you, I think I can make bigger tiles for detection but just want to know the limit of input size.

@erack1 3 жыл бұрын

New to machine learning and I'm wanting to create an object detection for video games. What are some good resources to start learning, I know the basics essentially of neural networks and their functions. I've bought your course and will be starting to learn that.

@muhammadjamil8171 3 жыл бұрын

Great content 😊 Thanks Sir !

@zubairsk1624 3 жыл бұрын

hello dear i hope you are okay i want to ask you few questions 1- can i apply some edit on yolo equation to get better detection 2- can you recommend me some videos that explain every thing about YOLOv4 3- how can i write these equations in python? i hope you answer me thank you

@jessmendoza1483 2 жыл бұрын

i've read some articles where they improve yolov3 by adding an equation, you should search some, maybe it could help you

@abdulwarissherzad9914 3 жыл бұрын

Nice, can we use it for YOLO object detector? If not or yes what is the reasons. Thank you ...

@valentynsichkar 3 жыл бұрын

Hello there! The Confusion Matrix displays mis-Classifications among classes. Any detection algorithm, after locating object on the image, has classification phase. Therefore, Confusion Matrix also can be build at this particular stage. The other case can be when ensemble of NNs are applied, e.g., one for detection and another for final classification.

@abdulwarissherzad9914 3 жыл бұрын

@@valentynsichkar Thank you for your prompt reply, I have already watched your previous video about the explanation of YOLOv3, so YOLOv3 or YOLOv4 when we run the mAP command line, it just calculate the TP and FP condition not another condition like TN, and FN, but at all it doesn’t have TN. How to calculate the confusion table without these four conditions, which we don’t have in YOLO value for this four condition. But these four conditions are important to have them exactly for each class that you want to classify and adding the value to confusion matrix.

@valentynsichkar 3 жыл бұрын

Yes, there are tools to help to calculate different metrics for YOLO, including Confusion Matrix. Have a look on GitHub by following keywords: "confusion matrix YOLO". Another one with more results: "YOLO metrics".

@abdulwarissherzad9914 3 жыл бұрын

@@valentynsichkar Thank you, the references were great, but I want to find out for "YOLOv4 custom object detector" a proper source code to count and print confusion matrix. Those references are for the coco dataset which is already trained by YOLOs authors. would you like to make a video for YOLO and SSD object detector about its mAP and Confusion matrix, because in recent years these two object detection algorithms have become popular.

@valentynsichkar 3 жыл бұрын

Thank you for the suggestion. I'll think about creating separate video lecture on how to compute Confusion Matrix for YOLO.

@glowwell4292 3 жыл бұрын

Thanks a lot. Explained neatly. Please make videos on V4 and V5 too.

@mohssineserraji1098 3 жыл бұрын

Great presentation

@sachinbharadwajm2120 3 жыл бұрын

great explanation & presentation!!!

@syafiqbasri8789 3 жыл бұрын

thank you so much sir.Its very useful and great explanation!

@listenbyheart5552 3 жыл бұрын

really awesome explanation it was! thanks a lot

@apurbaroy8411 3 жыл бұрын

Is it possible to integrate the YOLO algorithm with arduino or raspberry pi using a webcam?

@SM--wb4vg 3 жыл бұрын

Very well explained

@Alpha-hj2ss 3 жыл бұрын

Great Video! Can you please come with more videos

@shannondoyle5143 3 жыл бұрын

Really great detailed explanation. I don't get exactly what the ground truth values are determined for grid cells close to the centre grid cell of an object. Would you be able to explain this ?

@sumitbali9194 3 жыл бұрын

I have seen lot of videos on CNN, mostly crap. But your video is a gem. Appreciate the effort you have put into making this video. Diagrams are a great help in understanding the architecture. Thanks again

@pascalschluchter209 3 жыл бұрын

Hey, can someone explain to me, why the detection is happening in Layer 82, 94 and 106. Is there any mathmatical background or is it like a fix parameter of YOLOv3?

@m5a1stuart83 3 жыл бұрын

I was code in YoloV3 from Indian KZbinr, and now here I am learning the true nature of Yolo. It helps alot for this OCR Project where I can ignore the image that did not intended to be uploaded to Server.

@rlb5261 3 жыл бұрын

Thank. It is excellent!

@fatiah541 3 жыл бұрын

Thanks 🌹🌹🌹🌹

@fatiah541 3 жыл бұрын

🍀🍀🍀🍀🍀🇮🇶

@adithyanarayan9976 3 жыл бұрын

Spent multiple hours trying to read through various papers in order to understand some of the topics. Should've stumbled upon your channel and the video much earlier. Love the fact that everything is explained to the point. You've earned yourself a subscriber in me. Can't stress this enough, but please put out more videos like these, along the lines of Computer Vision. Well done mate and once again, THANK YOU SO MUCH!

@valentynsichkar 3 жыл бұрын

Thank you for the feedback! Will do!

@Stilbrech3rin 3 жыл бұрын

I can just follow the others. This video is very helpful. Did you publish a paper? I would like to cite you for my project.

@abdshomad 3 жыл бұрын

~ Timeline for watching again later ~ 00:01 Intro 01:17 What is YOLO? 03:13 Architecture of YOLO v3 05:28 Input 07:27 Detections at 3 Scales 09:28 Detection Kernels 12:02 Grid Cells 14:23 Anchor Boxes 18:25 Predicted Bounding Boxes 21:41 Objectness Score Conclusion

@bakervhaigaming9746 3 жыл бұрын

I regret why I haven't found this gem earlier! I had to go through 5-6 papers and hours of reading to understand these topics but your video made it very clear and specific. Please make more quality content like this. Thanks a lot.

@valentynsichkar 3 жыл бұрын

Thank you for the feedback! Will do!