Fewer predictions and lower score with current yolov5 version

[hannesoehler]

❔Question

I realized that with the current yolov5 version I get fewer predictions and a lower score compared to older versions of the repo (I think my last old version is from about 3 weeks ago). I ran detect.py with the same model and the same setting using the old and current version.

I attach the two outputs for the first images:

Any idea what could cause the difference? Did any default options for detect.py change lately? Maybe I should also mention that this is not due to TTA as I saw a change there in the last weeks.

[github-actions]

👋 Hello @hannesoehler, 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://www.ultralytics.com or email Glenn Jocher at glenn.jocher@ultralytics.com.

Requirements

Python 3.8 or later with all requirements.txt dependencies installed, including torch>=1.7. To install run:

$ pip install -r requirements.txt

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):

• Google Colab and Kaggle 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 (train.py), testing (test.py), inference (detect.py) and export (export.py) on MacOS, Windows, and Ubuntu every 24 hours and on every commit.

[glenn-jocher]

@hannesoehler there are many periodic changes to the repo code, and sometimes to the pretrained models, I would recommend you clone a clean version of the repo and retrain your model.

One performance impacting bug was introduced last week (and fixed a few days later), if you see a significant drop in performance this may be the cause.

Also, anecdotal results (a few examples) are not useful when comparing runs, you want to compare mAP on a test set.

[hannesoehler]

@glenn-jocher Thanks for your reply! The difference in mAP on the test set is not large (0.244 vs. 0.248).

Ok I will retrain the model with the current version and have a look at the results.

[hannesoehler]

I think I now know the cause: #2252