Thank you for your tutorials. They are very practical for me as I'm new to using AI models. I'd like to know if you've already made videos, or if you can make a video, on how to create models from scratch, especially techniques to achieve a good model that can capture the information we want. Let me give you an example: I want to recognize whether the bottle in my photo is a 2L Coca-Cola, 1L, or 600ml. Since they have very similar shapes and labels, pre-trained models like ResNet or others confuse them. If I create the model from scratch, I get worse performance. So I wonder, how do I push my model to focus on increasingly finer details to recognize the difference between these products?
In the mentioned scenario, you can work to get the size of the bottle. First detect the bottles using object detection model and then check the size of each bottle
from ultralytics import YOLO model = YOLO('yolov8.pt') for param in model.model.backbone.parameters(): param.requires_grad = False # Freeze backbone layers # If you want to freeze specific layers, you can do something like this: for name, param in model.named_parameters(): if "some_layer_name" in name: param.requires_grad = False # Replace with your specific layer names
This tutorial is tested on PC (64GB RAM, NVidia RTX 3090 24GB vRAM) but you can run AI programs (except LLMs) on any Laptop or PC which have minimum 8GB RAM and Nvidia's GPU with atleast 6GB vRAM. Processing speed will be slow as per the model you are testing but it will work.
The number of epochs needed for good results in object detection can vary widely based on several factors like for larger and more complex datasets needs more epochs. But there are some models which converge faster than others so in this case less epochs will also give you good results. Also, well-tuned learning rate can influence how quickly a model learns. Generally, many object detection models may start to show good results between 50 to 200 epochs.
@@CodeWithAarohi thx, i have just run your video training with the 20.000+ images and I noticed that after 80 epocs no major improvements have been donne. I did the same on hte 40.000+ dataset and I have seen the same.
Hi, have u tried to export model to onnx format, why its size is large best.pt = 45.2Mb, but my onnx model = 87.Mb.... I set format='onnx', dynamic=True, half=True, device=0(GPU), int8=True
Can we used semi-supervised model to train yolo like you do on custom dataset ? If yes, can you explain or do a video on this topic? Thanks a lot for your videos !
Hi, great video! I'm interested in learning how to export and run the YOLO11 model in TFLite format. Could you please share a tutorial or code snippet on this in your next video or as a comment? Thanks!
Copy move forgery detection in vedio using machine learning Use model: yolo Data set casia Please mam tell me how much time you upload vedio on above topic
I have done the exact same steps but after training I used results=model.val(data='testData.yaml', split='test') code to see the metrics of model on test data set. Is that correct use ?
Thanks sister, you nailed it 🔥 . I wanna know that can we use google colab for do this project? Because we don't need to have much gpu if we use google colab
Yes, You can use google colab. Code will be same. You only need to select the gpu from notebook settings of Colab and then just change the paths where ever required.
Hi, Amazing content right here! A video about evaluating the model and fine tuning would be amazing! Thanks again
Again. Worth watching your videos
Thank you so much 😀
one of the best video on yolo model. thank you.
Glad it was helpful!
thanks very clear, you are top teacher
Thank you!
better and better a pleasure this tutorial
Thank you!
Excellent work
Many thanks
Very well explained
Glad it was helpful!
Amazing video
Thanks!
Superb 👍👍👌👌👏👏
Thanks 🤗
Can you make a little advance, like first detecting the number plate and then extracing text from it.?
Thank you for your tutorials. They are very practical for me as I'm new to using AI models. I'd like to know if you've already made videos, or if you can make a video, on how to create models from scratch, especially techniques to achieve a good model that can capture the information we want. Let me give you an example: I want to recognize whether the bottle in my photo is a 2L Coca-Cola, 1L, or 600ml. Since they have very similar shapes and labels, pre-trained models like ResNet or others confuse them. If I create the model from scratch, I get worse performance. So I wonder, how do I push my model to focus on increasingly finer details to recognize the difference between these products?
In the mentioned scenario, you can work to get the size of the bottle. First detect the bottles using object detection model and then check the size of each bottle
Thank you for the clear instructions. Is there a way in this to freeze certain layers while fine tuning or retraining?
from ultralytics import YOLO
model = YOLO('yolov8.pt')
for param in model.model.backbone.parameters():
param.requires_grad = False # Freeze backbone layers
# If you want to freeze specific layers, you can do something like this:
for name, param in model.named_parameters():
if "some_layer_name" in name:
param.requires_grad = False # Replace with your specific layer names
@@CodeWithAarohi Thank you so much!
lovely vid ! cheers
Many thanks!
Amazing
Thanks
Thank you. I following your AI videos regularly. May i know which laptop u using for ML and DL models. It will help to practice to learning.
This tutorial is tested on PC (64GB RAM, NVidia RTX 3090 24GB vRAM) but you can run AI programs (except LLMs) on any Laptop or PC which have minimum 8GB RAM and Nvidia's GPU with atleast 6GB vRAM. Processing speed will be slow as per the model you are testing but it will work.
Thank you🙏
Thank you Madam
genia un beso , desde argentina
Thanks!
I trained v11 on rock paper scissor dataset, I am able to display the labeled image correctly but how can I get the predicted class ?
Yes plz ma'am
Thank you ❤
You're welcome 😊
based on your experience how many epocs give good results so that more than that it is a waste of time/resources?
The number of epochs needed for good results in object detection can vary widely based on several factors like for larger and more complex datasets needs more epochs. But there are some models which converge faster than others so in this case less epochs will also give you good results.
Also, well-tuned learning rate can influence how quickly a model learns.
Generally, many object detection models may start to show good results between 50 to 200 epochs.
@@CodeWithAarohi thx, i have just run your video training with the 20.000+ images and I noticed that after 80 epocs no major improvements have been donne. I did the same on hte 40.000+ dataset and I have seen the same.
Hi, have u tried to export model to onnx format, why its size is large best.pt = 45.2Mb, but my onnx model = 87.Mb.... I set format='onnx', dynamic=True, half=True, device=0(GPU), int8=True
mam im doing the same but for fire detection, how to print 'fire detected' in output from live cam only if fire detected . is it possible
Yes, you can do that. You just need to print "fire detected" if fire is detected. You can use if condition.
Can we used semi-supervised model to train yolo like you do on custom dataset ? If yes, can you explain or do a video on this topic?
Thanks a lot for your videos !
Hi, great video! I'm interested in learning how to export and run the YOLO11 model in TFLite format. Could you please share a tutorial or code snippet on this in your next video or as a comment? Thanks!
I haven't tried that yet but topic noted. I will definitely make video on requested topic.
Thank you Ma'am
Most welcome 😊
Copy move forgery detection in vedio using machine learning
Use model: yolo
Data set casia
Please mam tell me how much time you upload vedio on above topic
we can also use Roboflow to our own dataset and annotation?
Yes
I have done the exact same steps but after training I used results=model.val(data='testData.yaml', split='test') code to see the metrics of model on test data set. Is that correct use ?
docs.ultralytics.com/modes/val/
Thanks sister, you nailed it 🔥 . I wanna know that can we use google colab for do this project?
Because we don't need to have much gpu if we use google colab
Yes, You can use google colab. Code will be same. You only need to select the gpu from notebook settings of Colab and then just change the paths where ever required.
Thank you. Can I implement YOLO11 on jetson TX2?
You need python 3.6 or higher
Mam can you make a video on Conversational Image Recognition Chatbot. Please it would helpful..
I will try
please can u make a video about adding an interface like gradio huggingface to this model
Noted!
@CodeWithAarohi not uploaded yet 🤔😔
Can you please make a video without using API