code for the yaml file

[rmarkovic00]

Search before asking

• I have searched the YOLOv5 issues and discussions and found no similar questions.

Question

Hi, I'm trying to pre-train the yolov5m model downloaded from the ultralitics site on my custom dataset which has 3 classes: car,truck person. Is it necessary to match the class ID with the class ID of the coco dataset in yaml file? For example:

yaml file:
train: /path_to_train_images/
val: /path_to_val_images/

nc: 8 # Broj klasa od 0 do 7 (moraš uključiti sve klase do najvećeg ID-a!)

names: ["Person", "Unknown", "Car", "Unknown", "Unknown", "Unknown", "Unknown", "Truck"]

or its just okej like this:

yaml file:
train: /path_to_train_images/
val: /path_to_val_images/

nc: 8

names: ["Person", "Unknown", "Car", "Unknown", "Unknown", "Unknown", "Unknown", "Truck"]

because in coco dataset person has ID=0 car=2 and truck=7.

Additional

No response

[UltralyticsAssistant]

👋 Hello @rmarkovic00, thank you for your interest in YOLOv5 🚀 and for your question regarding custom dataset configuration!

It looks like you're trying to set up your YAML file for training YOLOv5 on a custom dataset. Here's some general guidance:

• For custom training questions, it's important to ensure your dataset and YAML configuration are properly set up. The names in the YAML file should correspond to the classes in your dataset, and their indices (IDs) should match the order in which they appear in the names list, rather than aligning with COCO class IDs. YOLOv5 does not require any alignment with COCO for custom datasets. Please ensure your dataset annotations are consistent with your names list.

If this is a 🐛 Bug Report, we kindly ask you to provide a minimum reproducible example (MRE), including details on reproducible YAML configurations, relevant code snippets, debugging logs, and any other information to help us diagnose the issue more effectively 🚀.

Additional Resources

Please also refer to our ⭐️ Tutorials for more guidance:

• Learn all about Custom Data Training.
• Avoid common pitfalls with these Tips for Best Training Results.

Requirements

Your environment should meet the following:

• Python>=3.8.0 installed with all requirements.txt dependencies, including PyTorch>=1.8.

To verify installation, run:

git clone https://github.com/ultralytics/yolov5  # clone
cd yolov5
pip install -r requirements.txt  # install

Verified Environments

YOLOv5 supports the following environments for training and running models:

• Free GPU-enabled notebooks:
  
• Cloud platforms with preconfigured environments:
  • See the GCP Quickstart Guide for Google Cloud.
  • Follow the AWS Quickstart for Amazon's Deep Learning AMI.
• Fully containerized training with Docker Quickstart Guide .

Status

To check current CI status, click below:

If the badge is green, all YOLOv5 Continuous Integration (CI) tests are passing ✅.

An Ultralytics engineer will assist you as soon as possible. Let us know if you have additional questions in the meantime! 😊

[rmarkovic00]

Thank you very much!

[sriram-dsl]

@rmarkovic00 When creating a custom dataset, there is no need to align your custom classes with COCO classes.

You can define your own class names and structure, such as:

nc: 3  
names: [person, car, truck]  # or [car, person, truck] or [truck, person, car] – the order doesn't matter  

However, you must ensure that the labeling data is correctly aligned with the class names in your dataset.

For example, if your labels are:

0: person  
1: car  
2: truck  

Then, the names should be:

names: [person, car, truck]

Any CNN or ML model does not directly deal with class names—it only uses them for one-hot encoding and for human readability.