github-actions[bot]
commented
1 year ago 👋 Hello @paolocomis, 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 screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.
If this is a custom training ❓ Question, please provide as much information as possible, including dataset images, training logs, screenshots, and a public link to online W&B logging if available.
For business inquiries or professional support requests please visit https://ultralytics.com or email support@ultralytics.com.
Requirements
Python>=3.7.0 with all requirements.txt installed including PyTorch>=1.7. To get started:
git clone https://github.com/ultralytics/yolov5 # clone
cd yolov5
pip install -r requirements.txt # installEnvironments
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.
glenn-jocher
paolocomis
lpc-eol
I've searched everywhere but haven't found a solution to my problem.
I trained a yolov5l6 model for the recognition of a very small tennis ball from a data set of about 16k labeled images (each image presents 1 or 2 balls) at 1280x720 resolution, obtaining a satisfactory recognition result, both when the ball is large enough (close to the camera) and when it is very small (away from the camera).
So I decided to also add player recognition by labeling the same images that now show 3 or 4 players and 1 or 2 tennis balls, with the exact same settings of the first training session. The end result is that player recognition is great, while the ball is now not detected as accurately as before, i.e. it is now often not detected when it is small enough (far from the camera).
I carried out the tests with the exact same videos and I realized that the confidence value of the ball has also dropped. If at frame X the ball confidence value was 0.87 with the new model the value dropped to 0.80.
I suspect it depends on the anchor settings. In fact, I noticed that if I train the model with only the annotations of the ball during the start-up phase, an unsuitable set of anchors is detected and therefore automatically fixed. When I throw the balls and players training on the other hand the anchor setup is judged ok.
Is it perhaps because there are more "players" objects than "balls" in the annotations and therefore the average player size is prioritized?
The training command I run on Colab is as follows:
python train.py --cache ram --epochs 350 --img 1280 --rect --batch 24 --worker 24 --weights yolov5l6.pt --cfg models/yolov5l6_custom.yaml --hyp data/hyps/hyp. scratch-high.yaml --data data/custom.yaml --name test --exist-okShould I perhaps adapt the training with a different setting than the standard one to solve the problem?
Thanks in advance