Feat:Train Yolov8n model

[Rishab87]

Describe the feature

To classify images in more categories we need to train yolov8n model as mentioned in PR #245, I tried training this model with model.train method of ultralytics but was not able to find a suitable dataset with bounding labels. So need to find a suitable dataset and train it on that to add more classes

Add ScreenShots

No screenshots to attatch

Record

• I agree to follow this project's Code of Conduct
• I want to work on this issue

[Rishab87]

If I find any dataset I'll train and create PR for it, meanwhile if anyone finds dataset and want to collaborate on this do tag me!

[kamisama-coder]

you only want dataset or the help in setting model

[Rishab87]

I only want the dataset

[Priyansh-max]

The yolo model is pre trained on 80 classes u need more than that?

[Rishab87]

yes to categorise images on more than those classes like screenshots, nature etc

[ShivanshuVerma83]

@Rishab87 I would like to work on this issue and contribute to expanding the classification capabilities of the YOLOv8n model.

As mentioned in the issue description, finding a suitable dataset with bounding box labels is crucial for training the model to classify images into more categories. I currently have access to a traffic dataset that includes annotations for various classes such as:

Vehicles (cars, trucks, buses, motorcycles, etc.)
Pedestrians
Other relevant objects (e.g., traffic lights, road signs, bicycles, etc.)
This dataset seems well-suited for training the YOLOv8n model to detect and classify objects in traffic scenarios. It also aligns with the goal of adding more classes to the model.

[Rishab87]

The classes which you mentioned here, this model is already trained on those classes, were you able to find dataset with other classes than these? (for eg screenshots, plants etc). If yes please do tell

[ShivanshuVerma8]

There are datasets which include more than 20000 categories like imagenet and big detection. In the dataset I mentioned above I can make a simulation based one and include categories like plants and all .

[Rishab87]

Just to confirm do they contain bounding labels?

[ShivanshuVerma83]

Yes they have labels too. I have spent considerable time on this and gone through research papers and found some notable datasets with labels and bounding boxes and other image annotations that classify objects .

[Rishab87]

If you want to train with these datasets then you can go ahead otherwise you can link those datasets here, I can train them I've evrything setup for it.

[ShivanshuVerma83]

I will train them too , and also i will link the datasets here :

1. OpenImage

2. Subset of this : https://github.com/EscVM/OIDv4_ToolKit?spm=5aebb161.5e6c11a6.0.0.1442c921jjFpz5
   Conversion Tool : Roboflow to filter and convert annotations.

3. VisDrone
   Already in YOLO format . No need for conversion

There is more please confirm if you need something like this .

[Rishab87]

I think this would work!

[ShivanshuVerma83]

@Rishab87 I have already written a script to fetch the model and have implemented some initial code as a starting point. Once the training process is complete, I would be more than happy to collaborate further and contribute meaningfully to the project. I look forward to working together and helping building this feature.