Yolo10

Community Help
1

Hi all,

I wish to apply a custom-training with YoloV10n on a dataset I have (I’ve been using YoloV8 until now).

Can you please elaborate whether I shall use from ultralytics import YOLOv10 or from ultralytics import YOLO? Does it matter? Does same answer hold for both training, inference and export (tflite, etc.)?

Regardless, is running a training for N epochs equivalent for running it for N/2 epochs followed by additional fine-tuning of N/2 epochs?

I’ve initially tried to use YOLOv10, and was able to complete successfully custom training and then inference. However I wasn’t able to export the model to tflite due to inconsistency assertions. I then switched back to YOLO (starting from a pre-trained yolov10.pt model) and was able to export it for tflite. Moreover, the official documentations of Ultralytics (link) do guide for using YOLO (and not YOLOv10) for YoloV10 training. On the other side RoboFlow tutorial (link) does guide to use YOLOv10… :thinking:

Thanks ahead,
DanitTech

2

Hello there! You can train using the repository from the creators of YOLOv10 where the model was launched, or the integration with Ultralytics. We recommend the Ultralytics approach and walk through it in our YOLOv10 training notebook? Both implement the same architecture, although there may be a few differences in implementations.

If you were not able to export the model from Ultralytics to TFlite, you may need to file an Issue on Ultralytics’ repository to ask if TFlite export is fully supported. Roboflow does not have weight conversion utilities for YOLOv10 model conversions.

2 Likes