Zero recall and zero precision even after 100 epochs and pretrained weights

[Yehor-Kovalenko]

Search before asking

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

Question

Hi!
I tried to train yolov5s model on my custom dataset, which was coorectly labeled with correctly created yaml file.
However when i try to train on custom dataset, even on very small one with only 10-20 images i get 0 precision and 0 recall values, just straight lines. Loss function however oscilates from 0 to some small value around 0.03.

Issue persists when I use pretrained weights and also when I use "zero" weights --weights '' even with 100 epochs

I experimented a bit and found if I copy the whole coco128.yaml into my custom_dataset.yaml and just change the first label to my label, and also add 10 my images to oficial coco128 dataset, then even after 1 epoch with pretrained weights model detects my object as a result of training.

How to fix this issue, so that model will work with my custom dataset and without creating the Frankestein of dataset?

Thank you in advance

Here is the results file of 50 epochs with only 1 image in the dataset using no pretrained weights, the same results when I use pretrained weights

Additional

No response

[github-actions]

👋 Hello @Yehor-Kovalenko, thank you for your interest in YOLOv5 🚀! Please visit our ⭐️ Tutorials to get started, where you can find quickstart guides for simple tasks like Custom Data Training all the way to advanced concepts like Hyperparameter Evolution.

If this is a 🐛 Bug Report, please provide a minimum reproducible example to help us debug it.

If this is a custom training ❓ Question, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Requirements

Python>=3.8.0 with all requirements.txt installed including PyTorch>=1.8. To get started:

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

Environments

YOLOv5 may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all YOLOv5 GitHub Actions Continuous Integration (CI) tests are currently passing. CI tests verify correct operation of YOLOv5 training, validation, inference, export and benchmarks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

Introducing YOLOv8 🚀

We're excited to announce the launch of our latest state-of-the-art (SOTA) object detection model for 2023 - YOLOv8 🚀!

Designed to be fast, accurate, and easy to use, YOLOv8 is an ideal choice for a wide range of object detection, image segmentation and image classification tasks. With YOLOv8, you'll be able to quickly and accurately detect objects in real-time, streamline your workflows, and achieve new levels of accuracy in your projects.

Check out our YOLOv8 Docs for details and get started with:

pip install ultralytics

[glenn-jocher]

Hello! 😊 It sounds like you're facing a tricky training issue. Zero precision and recall usually suggest that the model isn't learning to generalize from your data effectively. Here are a couple of suggestions:

1. Dataset Size and Variety: The size of the dataset could be very crucial, especially for deep learning models. A tiny dataset of 10-20 images might not provide enough variability for the model to learn effectively. Try increasing the size of your dataset if possible.

2. Hyperparameters: Adjusting hyperparameters like the learning rate or augmentation strategies may also help. Experimenting with these could uncover a more optimal training configuration.

3. Validate Labels and Annotations: Ensure that labels are indeed correct and properly formatted. Double-check paths and formats in your YAML file.

4. Baseline Check: Start with training on a smaller, proven dataset like coco128 to verify everything functions properly without any modifications. This can help isolate whether the issue is with the dataset or the model configuration.

5. Debug Output: Utilize the --debug flag during training to get more detailed output that might point to what's going wrong.

If these steps don't lead to improvements, please provide further details about your training configuration and any modifications you made to the codebase. This information can help diagnose the issue more effectively. Keep at it! You're on the right track by experimenting and scrutinizing your approach. 👍