YOLOR on a Custom Dataset | Object detection using YOLOR

I will show you how to perform object detection using YOLOR which stands for You Only Learn One Representation.
Github: github.com/Aar...
Yolor is a new object detection algorithm and released in may 2021,
If you will see the performance comparison of yoloR with other models in their official paper. You will get to know that yoloR is the fastest object detection model .
YoloR is written in both (tensorflow & torch). Today I will show you how to run yolor using pytorch
Yolor has different model variations including, yolor-p6(small), yolor-w6(medium), yolor-d6(large), yolor-e6(extra-large)
If you have any queries, you can comment or you can mail me at [e-mail]
Join this channel to get access to perks:
/ @codewithaarohi
#YOLOR
#ObjectDetection
#Yolo
#ComputerVision

[@soufianechehboune4811]

I would say just one thing : Thank you Aarouhi you save my life

[@CodeWithAarohi]

Glad my video helped you!

[@soufianechehboune4811]

@@CodeWithAarohi actually all your videos helps me to implement all different series of YOLO... I can't found words my gratitude to you! Thank you so much again 💗😊

[@CodeWithAarohi]

@@soufianechehboune4811 Glad to hear that 😊

[@preethamdp729]

your github link throwing 404 error

[@CodeWithAarohi]

Try this github.com/AarohiSingla/Object-Detection-Using-YOLOR

[@kishoreg8835]

mam you made a lot of object detection tutorials. but which is best for real life project which can be made into a product for a company to sell. it should be very very fast and accurate. you have to reply to this question mam. it's a genuine question.

[@CodeWithAarohi]

As you said- I have to reply to this question, So here is my reply - It always depend upon the type of industry (Medical, agriculture, security etc) where you are using this Object detection. Also the resources which you are using to run the Object detection. And then check if task is to detect smaller objects or larger objects. These kind of parameters are helpful before choosing a suitable Object detection model. I will try to make a video which will explain this topic in detail :)

[@MrIWANTCHICKEN]

Easy tutorial. Is it possible to view the progress of the training eg: like chart/graph in yolor? what is the method to do that

[@SuperJude67]

Great tutorial, you are awesome lecturer

[@CodeWithAarohi]

Glad it was helpful!

[@utkarshtripathi9118]

Ossm, Exellent, Best of the Bset

[@AnjanaSivakumar]

Thank you so much. Could you please do the same using YOLOX as well?
Also please try to do it using google colab, that would be more helpful

[@CodeWithAarohi]

Will try

[@AnjanaSivakumar]

@@CodeWithAarohi @Code With Aarohi Thank you ma'am. Is there any way I can contact you? I need some help for my MTech project.

[@colorless5006]

thanks for your effort and time. which annotation software is used? does it matter which annotation tool should be used?

[@CodeWithAarohi]

I took this dataset from kaggle but you can annotate your images using labelImg tool

[@haminlim5272]

Your videos are amazing! Is there a way to do the real-time detection? Especially with the rtsp uri.
Thank you so much!

[@CodeWithAarohi]

I think we can use it like --source rtsp://111.123.122.1:5050/stream1 but not sure

[@haminlim5272]

@@CodeWithAarohi Thank you so much! I also watched your Yolov7 official version and trying to work on it. I will leave a comment on that video If I have any questions!

[@pifordtechnologiespvtltd5698]

Keep it up 👍

[@CodeWithAarohi]

Thanks!

[@preethamdp729]

does Slicing Aided Hyper Inference (SAHI) library supports yoloR?

[@akbaraliotakhanov1221]

Nice tutorial! Thanks for your effort. By the way github link is not found, could you reshare the link?

[@CodeWithAarohi]

github.com/AarohiSingla/Object-Detection-Using-YOLOR

[@eranfeit]

Hi,
Can you share a link to the traffic dataset ?
Eran

[@CodeWithAarohi]

www.kaggle.com/datasets/valentynsichkar/traffic-signs-dataset-in-yolo-format

[@nayoung32918]

What is the difference between batch, width, height, and --batch-size, --img of the train command in the cfg file?

[@CodeWithAarohi]

This is the train command and it has batch size and image: (batch size is the size of the batch and -img is the dimension of the image) python train.py --batch-size 8 --img 1280 1280 --data custom.yaml --cfg cfg/yolor_p6_custom.cfg --weights '' --device 0 --name yolor_p6 --hyp hyp.scratch.1280.yaml --epochs 300

[@chrisber]

Hey, thanks so much, this tutorial was EPIC!
I'm trying to run my custom YOLOR on custom images which aren't square. My first question is, will I have to change the width and height in the .cfg file? I presumed that I would.
So I changed the width and height in the .cfg file and then trained a model.
So now that I just trained a model on non-square images, each image was 1952 wide by 960 high (each dimension is divisible by 32).
My second question is; when running my trained model from the command line, how do I specify the second dimension parameter? As it only wants to accept one for some reason (implying that all images must be square). I keep getting the error "detect.py: error: unrecognized arguments: 960"
Thanks for any help :)

[@CodeWithAarohi]

Can you share the command you are using

[@omaramini1670]

Thanks for your effort. By the way github link is not found, could you reshare the link?

[@CodeWithAarohi]

github.com/AarohiSingla/Object-Detection-Using-YOLOR

[@utkarshtripathi9118]

Mam,
could you pleas make a video on how to deploy our model on web server .
If Already is there any video pls Send me the link...........

[@CodeWithAarohi]

th-cam.com/video/_Iq1P1Wz_RU/w-d-xo.html

[@amanharis1845]

Ma'am can we run this program to detect objects on real time video ?

[@warrior_1309]

Just try to convert the model into onnx format and read it through opencv for real time inferencing.

[@Kishi1969]

Please How to check my GPU in my window 10 please?

[@CodeWithAarohi]

You ca check from Task Manager in windows

[@Kishi1969]

@@CodeWithAarohi Thanks

[@hariomdubey03]

Nice video,
Thanks

[@CodeWithAarohi]

Welcome

[@Kishi1969]

Dear Ma, If i'm using classes like 6, means (classes + 5)= (6+5)3 right ? 33 right in filters?

[@CodeWithAarohi]

Correct

[@Kishi1969]

@@CodeWithAarohi Thanks

[@danb9393]

I did same all the things that you did. but my model is not detect bouding box? what is the problem?

[@CodeWithAarohi]

There are various reasons: 1-Maybe your model need more training. 2- Dataset is not proper. 3- Try to tune hyper parameters etc

[@danb9393]

@@CodeWithAarohi yes you right.. Thank you. When I did more training epoch I gave me detection. Thank you. your series very clear. Please make a video about YOLOX_custom_data too.

[@Swearinbag]

Is there a reason why we have to adjust the filters so specifically? :) Might be a stupid questions but I've seen this pop up quite a lot but can't seem to find a source as to why this happens...

[@CodeWithAarohi]

Number of filters are decided by this method : (classes+5)3 . Suppose you have 4 classes then = (4+5)3 = 27

[@Swearinbag]

@@CodeWithAarohi Oh I understood that! But I meant why you add the 5 and then multiply the whole thing by 3

[@darkmember727]

@@Swearinbag Detection is performed on 3 anchor boxes, and gives 5 output: cx, cy, w,h, prob.score + n classes

[@animators4126]

can you please show me one of your label file, I converted mine with same name as image where I am keeping bbox values and class id but its not working

[@CodeWithAarohi]

Content of 1 txt file : 1 0.5841911764705883 0.535625 0.030147058823529412 0.04375

[@justusbieberstein6699]

What program do you use to annotate the images?

[@CodeWithAarohi]

LabelIMG tool github.com/tzutalin/labelImg

[@dariosantiago257]

do you know if i can run this model on jetson nano?

[@CodeWithAarohi]

I never tried it but I think it will work as we can run different types of Neural Networks on it.

[@RAZZKIRAN]

thank u madam,

[@CodeWithAarohi]

Welcome 😊

[@leandromontenegropinto6704]

Hey, nice video!! Do you have for google colab? Thanks

[@CodeWithAarohi]

Sorry, no

[@andrewh5640]

Very good

[@CodeWithAarohi]

Thanks