This is a really great initiative!! Thanks man! I hope you cover segmentation algorithms and their implementations in PyTorch as well! That would be a complete series!
Great work brother,,was eagerly waiting for this.i have been following ur videos for quite a long time an it has helped me in my journey of learning the amazing Pytorch framework and implementing deep neural network models from scratch.I requested you by commenting in one of your videos to come up with the object detection series and at present I am overwhelmed to see this playlist.Cant wait to get my hands dirty in implementing YOLO from scratch.kudos to u brother.
I remember you've expressed that object detection is a much needed series and thank you for the push! Hopefully I can do a good job on covering this area in depth, and also it will lay a foundation to future videos in object detection with perhaps more architectures etc.
Thanks bro! i love your video! That can help me understand about the object detection more easily! I am a beginner on deep learning from China ~ .Finally i believe we can be the good friends
Love this! Would you be covering other architectures in addition to Yolo? I've been trying to understand this stuff recently, but I can't picture what a single training cycle flow looks like for even relatively simpler architectures like overfeat. More than explaining the individual concepts, I am DYING to see how the pieces fit together to make just a single training cycle. I just can't seem to find a resource that does this, to help me break into the world of object detection. I hope you will turn out to be that resource I've been searching for. (sorry for the rant :))
My plan right now is not to, mostly because it is incredibly time consuming implementing these architectures from scratch, but if many people wants to see a particular architecture I'll do it. I can definitely relate to how you feel, my realization is that object detection is quite messy (contrary to image classification) and there's a lot of details that goes into a training cycle. But rest assured, I believe we will build a solid foundation in this area by first starting to understand the related pieces and then we will tackle the project of implementing yolo from scratch. I will show how to implement the model, loss function, data loading, etc. so afterwards you will (hopefully) feel that you have a thorough understanding of the concepts behind object detection (both theory and implementation in pytorch)!
If enough people want it yes, but I think the code is so similar that even if you use TF you can still be able to implement it from the video, or do you think that would be difficult?
@@AladdinPersson Now that I've been watching the series, I noticed it is more of a scratch implementation (im loving it so much btw) so yes, it should not be a big deal to implement in TF. What I mentioned in my comment was TF Object detection API, which is more high level, using pre-trained models etc. Thank You!!
Hey! Back again in your channel skipping tensorflow tutorials :3 However I would request you to push the object detection tutorials with explaining some SOTA architectures. If you please!
Hey, really enjoyed this video. Have you heared of FCOS object detector? Maybe you can consider to showcase this one in your next paper review? Best regards
@@zahidhossen3134 Oh, I guess I confused "points" with "nodes" in the video as he used them interchangeably. At least 4 nodes*, which describe 2 points*, (or only 1 point described by 2 nodes + width node + height node). Thanks!
some videos says that object localization means just locating the objects(can be multiple) but you are saying it means locating and labelling single object. Which one is correct?
Hey Aladdin, To begin with, your channel has some amazing content out there!!!!!🙌🙌 I appreciate all your diligent efforts and dedication.....👏👏👏 I am glad that I came across your channel!!! Already subscribed, liked and turned on my notification button!!!😃😃 After watching your videos, I have a doubt that's bothering me. I would like to know, how to deal with existing annotated dataset for a different object detection problem. For example: If I have a already annotated dataset annotated in PASCAL VOC with bounding boxes for (say cats, dogs and birds) found in an image. How can I perform detection of only cat and dogs by discarding bounding boxes/annotations of bird? It would be of great help if you could suggest me an idea or your approach to deal this problem. Thank you in advance. Keep shining and rocking!!!!🙌🙌👏🎉😊
Hm, yeah I was thinking after implementing Yolo from scratch to use some pretrained ones that we can train on some datasets. But maybe not faster RCNN, feels like there should exist a lot better ones?
hello i have a question why i use yolo to detect multiple objects in image what i think is to use a pretrained model like vgg16 and modify it’s ouput layer to have neuron for each class i have with a sigmoid activation to be able to classify multiple objects in image and have multiple neurons for cooredinates for each object so why we use yolo?
This was just an introduction to object detection but I will be showing how to implement the concepts in pytorch. Did you anything particular example in mind?
@@AladdinPersson implementing Feature Pyramid Network + RPN or SSD Net can be a good implementation. I think the feature extraction part is ok but bounding box regression can be challenging for me
Does anyone have any experience training a deep neural network? I am working on an autonomous robot project and not sure of a realistic way of collect images, classifying them, and training them
Collecting is totally up to you. I mean, if you wanna be so specific (like working for a company project), you have to get out and collect your data goes on with your style. Classifying is really simple actually. Just hire interns and tell them to use label tools. Training is where you show your skills actually.
Great! Looking forward to the YOLO implementation from scratch :)
This is a really great initiative!! Thanks man! I hope you cover segmentation algorithms and their implementations in PyTorch as well! That would be a complete series!
Hope for another great series. Thanks for sharing.
Dude I just waiting for this stuff.
Perfect timing ;) Will try to make the next video by tomorrow on evaluating bounding boxes with intersection over union & implementation in pytorch
Great work brother,,was eagerly waiting for this.i have been following ur videos for quite a long time an it has helped me in my journey of learning the amazing Pytorch framework and implementing deep neural network models from scratch.I requested you by commenting in one of your videos to come up with the object detection series and at present I am overwhelmed to see this playlist.Cant wait to get my hands dirty in implementing YOLO from scratch.kudos to u brother.
I remember you've expressed that object detection is a much needed series and thank you for the push! Hopefully I can do a good job on covering this area in depth, and also it will lay a foundation to future videos in object detection with perhaps more architectures etc.
Great job. Cleary explanation. Motivated me to done all chapters
Brother thanks for these computer vision videos! Love from India
Really helpful clear work. Thank you.
I was waiting for this series, hope to see more videos.
Lovely introduction: informative and pleasant to listen to :) Will dig deep into your channel now!
Great series. Look forward to your more series.
I'm glad I came across your video when I was confused about target detection😀
Very well explained!
Thanks man for your work
You are doing such an incredible join man
extremely good explanation
Awesome explainer thanks for making this content :) Didn't know how these classifiers worked and this video gave me a useful intuition
Thanks for this informative tutorial
Thank you for this amazing video it's very helpful :)
thank you, that was greatly explained
Thanks for this! It helps a lot (I'm doing an internship in computer vision!)
Great video, easy to follow and well structured graphics for visual learning
Thanks bro! i love your video! That can help me understand about the object detection more easily! I am a beginner on deep learning from China ~ .Finally i believe we can be the good friends
Thunder buddies for life ;)
@Aladdin Persson.... thanks again... amazing amazing... once again amazing.....
Really appreciate the positive energy & support 🙏🙏
Yo, couldn't wait to see upcoming videos, I'm sure you'll kill it.
🙏👊
Love this! Would you be covering other architectures in addition to Yolo? I've been trying to understand this stuff recently, but I can't picture what a single training cycle flow looks like for even relatively simpler architectures like overfeat. More than explaining the individual concepts, I am DYING to see how the pieces fit together to make just a single training cycle. I just can't seem to find a resource that does this, to help me break into the world of object detection. I hope you will turn out to be that resource I've been searching for. (sorry for the rant :))
My plan right now is not to, mostly because it is incredibly time consuming implementing these architectures from scratch, but if many people wants to see a particular architecture I'll do it. I can definitely relate to how you feel, my realization is that object detection is quite messy (contrary to image classification) and there's a lot of details that goes into a training cycle.
But rest assured, I believe we will build a solid foundation in this area by first starting to understand the related pieces and then we will tackle the project of implementing yolo from scratch. I will show how to implement the model, loss function, data loading, etc. so afterwards you will (hopefully) feel that you have a thorough understanding of the concepts behind object detection (both theory and implementation in pytorch)!
@@AladdinPersson yeah I can understand. Thanks for those reassuring words. Looking forward to the Yolo series!
wooow that was really amazing
That's amazing. Thank you! Are you planning to cover Tensorflow too, in a future (TF Object detection API)?
If enough people want it yes, but I think the code is so similar that even if you use TF you can still be able to implement it from the video, or do you think that would be difficult?
@@AladdinPersson Now that I've been watching the series, I noticed it is more of a scratch implementation (im loving it so much btw) so yes, it should not be a big deal to implement in TF. What I mentioned in my comment was TF Object detection API, which is more high level, using pre-trained models etc. Thank You!!
Hey! Back again in your channel skipping tensorflow tutorials :3 However I would request you to push the object detection tutorials with explaining some SOTA architectures. If you please!
Hey, really enjoyed this video. Have you heared of FCOS object detector? Maybe you can consider to showcase this one in your next paper review? Best regards
3:49 why at least 4 though? couldn't we have a center point + width + height? or is that not feasible by the CNN
Center point will have two numbers as coordinates.
@@zahidhossen3134 Oh, I guess I confused "points" with "nodes" in the video as he used them interchangeably. At least 4 nodes*, which describe 2 points*, (or only 1 point described by 2 nodes + width node + height node). Thanks!
Thanks for the video. Have yoou come any service or application for convolutional neural networks in remote sensing applications ?
some videos says that object localization means just locating the objects(can be multiple) but you are saying it means locating and labelling single object. Which one is correct?
OBD with tf please
Hey Aladdin,
To begin with, your channel has some amazing content out there!!!!!🙌🙌
I appreciate all your diligent efforts and dedication.....👏👏👏
I am glad that I came across your channel!!! Already subscribed, liked and turned on my notification button!!!😃😃
After watching your videos, I have a doubt that's bothering me. I would like to know, how to deal with existing annotated dataset for a different object detection problem.
For example: If I have a already annotated dataset annotated in PASCAL VOC with bounding boxes for (say cats, dogs and birds) found in an image.
How can I perform detection of only cat and dogs by discarding bounding boxes/annotations of bird?
It would be of great help if you could suggest me an idea or your approach to deal this problem.
Thank you in advance.
Keep shining and rocking!!!!🙌🙌👏🎉😊
very great
Awesome!
THANK you please can you make a video on yolo v5
Thanks~~~
There you go :)
NICE!
Great work! Can you demonstrate how to train built-in object detectors in pytorch like fasterRCNN(fasterrcnn_resnet50_fpn) on large datasets?
Hm, yeah I was thinking after implementing Yolo from scratch to use some pretrained ones that we can train on some datasets. But maybe not faster RCNN, feels like there should exist a lot better ones?
hello i have a question
why i use yolo to detect multiple objects in image what i think is to use a pretrained model like vgg16 and modify it’s ouput layer to have neuron for each class i have with a sigmoid activation to be able to classify multiple objects in image and have multiple neurons for cooredinates for each object so why we use yolo?
Because we don't know beforehand how many objects there will be in the image.
Great
Are you working professional or undergrad?
I'm a masters student
can you make examples in pytorch like you did before
This was just an introduction to object detection but I will be showing how to implement the concepts in pytorch. Did you anything particular example in mind?
@@AladdinPersson implementing Feature Pyramid Network + RPN or SSD Net can be a good implementation. I think the feature extraction part is ok but bounding box regression can be challenging for me
@@mustafabuyuk6425 My plan is to implement yolo from scratch, but if others want to see other architectures I'll look into it
all this time i read bboxes like bboy thinking it was just a nickname but instead theyre for bounding boxes
Does anyone have any experience training a deep neural network? I am working on an autonomous robot project and not sure of a realistic way of collect images, classifying them, and training them
Collecting is totally up to you. I mean, if you wanna be so specific (like working for a company project), you have to get out and collect your data goes on with your style. Classifying is really simple actually. Just hire interns and tell them to use label tools. Training is where you show your skills actually.
or if you dont give a f, then use kaggle sets etc.
Yolo = Bayesian learning + box detection