training actually can not work

[lucasjinreal]

After trained on several epochs, I found the mAP is still very low. Does the training really works?

    Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
    14/299      6.4G   0.02273  0.002925 0.0003764   0.02603        11       640: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 6960/6960 [54:20<00:00,  2.13it/s]
               Class      Images     Targets           P           R      [email protected]  [email protected]:.95: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████| 6960/6960 [13:37<00:00,  8.51it/s]
                 all    5.57e+04    1.74e+05    0.000332     0.00039     2.4e-06    8.59e-07

     Epoch   gpu_mem      GIoU       obj       cls     total   targets  img_size
    15/299      6.4G   0.02232  0.002874  0.000371   0.02556         7       640: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████████████████| 6960/6960 [54:36<00:00,  2.12it/s]
               Class      Images     Targets           P           R      [email protected]  [email protected]:.95: 100%|█████████████████████████████████████████████████████████████████████████████████████████████████████████████| 6960/6960 [14:23<00:00,  8.06it/s]
                 all    5.57e+04    1.74e+05    0.000342    0.000401    2.44e-06    8.66e-07

[github-actions]

Hello @jinfagang, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Google Colab Notebook, Docker Image, and GCP Quickstart Guide for example environments.

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 model or data training question, please note that Ultralytics does not provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as:

• Cloud-based AI surveillance systems operating on hundreds of HD video streams in realtime.
• Edge AI integrated into custom iOS and Android apps for realtime 30 FPS video inference.
• Custom data training, hyperparameter evolution, and model exportation to any destination.

For more information please visit https://www.ultralytics.com.

[glenn-jocher]

@jinfagang training operates correctly. See https://github.com/ultralytics/yolov5#reproduce-our-training

[lucasjinreal]

@glenn-jocher Why am I training so low AP

[glenn-jocher]

@jinfagang we've completed writing a custom data training tutorial here that may help. Make sure you check your labels by looking at train*.jpg after training starts.
https://docs.ultralytics.com/yolov5/tutorials/train_custom_data

[tangxiaochu123230]

could you find the problem,I have a similar situation

[annie7777]

had similar problem and just found that I forgot to change the number of class in yolov5s.yaml when following the tutoiral.

yolov5/models/yolov5s.yaml

Line 2 in bf6f415

nc: 80 # number of classes

[glenn-jocher]

@annie7777 thanks for the comment! Recent updates have actually removed this requirement. Class counts for training are now inherited from the dataset.yaml nc in case of a difference.

The code changes often, so if anyone runs into a similar bug git pull to update your code to simply clone a fresh copy of the repo!

[github-actions]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.