Ultralytics•5w ago

yolo model train error

Hi ,What should I pay attention to if I want to train a model that is not pre training? because i has try to train a few time,every time in the mAP does not go up. i try use30000 image or 60000image or 10000image,and use diff dataset,i just want to train 1 class thank you so much

33 Replies

Skillnoob_•5w ago

Are you trying to continue training a model? Because I see resume=True in there

h4mmer•5w ago

harry cowOP•5w ago

Yes, because sometimes I interrupt the training. I saw in online tutorials that by enabling this, I can resume training from where it was interrupted.

Skillnoob_•5w ago

If you want to continue a interrupted training that didn't finish, you can point YOLO() to the last.pt and use resume=True

harry cowOP•5w ago

Thank you. Most of my training sessions are not interrupted. May I ask how many photos I should use for training? My mAP never increases with each session.

Skillnoob_•5w ago

Can you share the labels.jpg and results.jpg from the train folder?

harry cowOP•5w ago

thank you so much

harry cowOP•5w ago

https://universe.roboflow.com/microsoft/coco i download the data from this site

Skillnoob_•5w ago

no not those there are files that are called labels.jpg and results.jpg in the train folder ultralytics creates

harry cowOP•5w ago

result

Skillnoob_•5w ago

They are image files called that inside the runs folder

harry cowOP•5w ago

i dont have file in runs folder

harry cowOP•5w ago

i change the path to D:/

Skillnoob_•5w ago

thats just windows not loading the folder

harry cowOP•5w ago

let me find it,i am sorry 🙏

Skillnoob_•5w ago

then it should be located in that folder

harry cowOP•5w ago

i gusee i dont have labels.jpg?

harry cowOP•5w ago

Skillnoob_•5w ago

Wait why are you training on the coco dataset The default models are trained on coco

harry cowOP•5w ago

because i try to train other dataset,but still failed. so i try cocodataset 😆 i want to test is it data problem😆

Skillnoob_•5w ago

Then you should say what the issue with your original dataset is What are you trying to train?

harry cowOP•5w ago

i trying to train ,detect to person . and i og dataset issue,is the mAP Not increasing,keey in 0.00xxx

Skillnoob_•5w ago

You can use the original coco models if you want to detect people

from ultralytics import YOLO

model = YOLO('yolo11n.pt')

model.predict(source='path/to/image.jpg', classes=[0])

from ultralytics import YOLO

model = YOLO('yolo11n.pt')

model.predict(source='path/to/image.jpg', classes=[0])

With this code you will only detect people

harry cowOP•5w ago

my professor said I cannot use a model that has already been pre-trained 🫠

Skillnoob_•5w ago

Amazing, great professor You can try:

from ultralytics import YOLO

model = YOLO('yolo11n.yaml')

model.train(data='coco.yaml', classes=[0], batch=-1)

from ultralytics import YOLO

model = YOLO('yolo11n.yaml')

model.train(data='coco.yaml', classes=[0], batch=-1)

This will train the model on the coco dataset with only the person class If you specify coco.yaml, ultralytics will automatically download the dataset it needs

harry cowOP•4w ago

thanks for your help🙏 ,but it already download few hour,is it normal ?

Skillnoob_•4w ago

Yes, coco is a large dataset The dataset is about 20gb, how fast it is depends on your internet speed

harry cowOP•4w ago

thank you,btw can i change the model to yolo12n.yaml?

Skillnoob_•4w ago

I wouldn't recommend yolo12, its worse than 11

harry cowOP•4w ago

May I ask what the reason is? From the data, it seems that 12 is faster and more accurate.

Skillnoob_•4w ago

Its much slower on cpu

Gaming

Programming

yolo model train error

Did you find this page helpful?