capitaletech/language-levels-yolov10-v1

language-levels-yolov10m

This repository contains the fine-tuning script and weights of YOLOv10m on languge levels dataset.

The language levels dataset :

• The dataset contains about 50000 images (some corrupt images were removed).

The training :

• We fine-tuned YOLOv10m with this configuration : Model: yolov10m.pt

  Epochs: 10

  Batch: 12

  device: [0,1] (GPU T4 x 2)

  size : {'width': 799, 'height': 151}

• Training time : Wall time: 3h 24min 42s

The results :

Confusion matrix :

Training loss :

Best Training Box loss: 2.0493 , on epoch: 10

Best Validation Box loss: 1.7033 , on epoch: 10

==================================================

Best Training Cls loss: 1.1808 , on epoch: 10

Best Validation Cls loss: 0.84118 , on epoch: 10

==================================================

Best Training DFL loss: 1.7253 , on epoch: 10

Best Validation DFL loss: 1.6617 , on epoch: 10

Precision, F1 and recall :