How to understand the result file

[BaolinWang98]

Shortly after I touched the target detection, I want to ask a silly question, what is the meaning of each column in result.txt?

[github-actions]

Hello @stonerwang, 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.

[glenn-jocher]

@stonerwang you can plot your results.txt file rather than looking at it like neo in the matrix. From https://github.com/ultralytics/yolov3#training:
from utils import utils; utils.plot_results()

The function has the column titles.

[BaolinWang98]

Thank you for your reply.
I can't find the weight file which named ultralytics68.pt in Google Cloud,Please help me.

Another question:If i use python train.py --weights ultralytics68.pt to training. Whether only the parameters of the daeknet will be initialized, and the parameters of the Yolo layer will be trained from 0 ？

Thanks again！

[glenn-jocher]

@stonerwang ultralytics68.pt is deprecated. It's replacement is yolov3-spp-ultralytics.pt.

FYI the repo automatically attempts to download this file automatically if specified, so you should not need to manually download from google drive.

[github-actions]

This issue is stale because it has been open 30 days with no activity. Remove Stale label or comment or this will be closed in 5 days.

[priteshgohil]

Result.txt with column index. Correct me if I'm wrong.

0.sr.no. 1.Epoch, 2. , 3.GIoU loss, 4.Objectness loss, 5.Classification loss, 6., 7., 8., 9.Precision, 10.Recall, 11.mAP@0.5, 12.F1, 13.Val GIoU, 14.Val objectness, 15.Val classification

[glenn-jocher]

@priteshgohil yes looks right. You can plot with utils.plot_results()

[EthanGD]

Sorry, So what is the meaning of each column in result.txt?

[EthanGD]

['epoch','train_GIOU_loss','train_obj_loss','train_cls_loss','total','target','img_size','precision','recall','MAP@0.5','F1','val_GIOU_loss','val_obj_loss','val_cls_loss']
is it right?

[Ying5775]

["Epoch", "GPU_Mem", "train_box_loss", "train_obj_loss", "train_class_loss", "train_total_loss", "labels", "img_size", "Precision", "Recall", "mAP@.5", "mAP@.5:.95", "val_box_loss", "val_obj_loss", "val_class_loss"]
I could be wrong

[glenn-jocher]

@Ying5775 The columns in the result.txt file correspond to:

1. Epoch
2. GPU_Mem
3. train_box_loss
4. train_obj_loss
5. train_class_loss
6. train_total_loss
7. labels
8. img_size
9. Precision
10. Recall
11. mAP@.5
12. mAP@.5:.95
13. val_box_loss
14. val_obj_loss
15. val_class_loss

[EricaMendonca1994]

There is no IoU?
Can I calculate these metrics?

[glenn-jocher]

You can calculate IoU using the validation mode in YOLOv3, which provides metrics like mAP that rely on IoU. For more details, check the YOLO performance metrics guide.