I trained the model with 2026, but detected 2 same boxes on one object

[fangzhengzhu]

Hello, when I predicted with the model trained with yolo26, 2 similar boxes with same label were on one area. According to NMS free and DFL Free, I think there should be only one exists.

And I also found DFL in the final model file named best.pt, Does it mean the final model is not really a yolo26 model?

[UltralyticsAssistant]

👋 Hello @fangzhengzhu, thank you for your interest in Ultralytics 🚀! This is an automated response to help get things moving, and an Ultralytics engineer will also assist you soon.

We recommend a visit to the Docs where you can find many Python and CLI usage examples, and where many of the most common questions may already be answered.

Since this appears to be a model behavior/question report, please share a minimum reproducible example so the team can better investigate 🔍. Helpful details include:
-Your exact training and prediction commands
-Your ultralytics version
-A small sample image or dataset example
-Training logs/results
-Any custom config changes you made

If this is related to custom training ❓, please provide as much information as possible, including dataset image examples and training logs, and verify you are following our Tips for Best Training Results.

Join the Ultralytics community where it suits you best. For real-time chat, head to Discord 🎧. Prefer in-depth discussions? Check out Discourse. Or dive into threads on our Subreddit to share knowledge with the community 💬.

Upgrade

Upgrade to the latest ultralytics package including all requirements in a Python>=3.8 environment with PyTorch>=1.8 to verify your issue is not already resolved in the latest version:

pip install -U ultralytics

Environments

YOLO may be run in any of the following up-to-date verified environments (with all dependencies including CUDA/CUDNN, Python and PyTorch preinstalled):

• Notebooks with free GPU:
• Google Cloud Deep Learning VM. See GCP Quickstart Guide
• Amazon Deep Learning AMI. See AWS Quickstart Guide
• Docker Image. See Docker Quickstart Guide

Status

If this badge is green, all Ultralytics CI tests are currently passing. CI tests verify correct operation of all YOLO Modes and Tasks on macOS, Windows, and Ubuntu every 24 hours and on every commit.

[fangzhengzhu]

ultralytics version: 8.4.33

train code:
`from ultralytics import YOLO

Load a pretrained YOLO26n model

model = YOLO("./model/yolo26m.pt")

Train the model on the COCO8 dataset for 100 epochs

train_results = model.train(
data="/home/swzx/source/OCRAutoScore/segmentation/Layout4Card/dataset/dataset.yaml", # Path to dataset configuration file
epochs=100, # Number of training epochs
imgsz=1024, # Image size for traininga
workers=4,
device=[0,1], # Device to run on (e.g., 'cpu', 0, [0,1,2,3])
)
`

predict code:
def predict(self, image_path): results = self.model(image_path, conf=0.5, iou=0.6, verbose=False) return results

partial dataset files:
label.zip

I didn't save the training logs.

No config changes I had made.

[glenn-jocher]

Thanks for the details — with YOLO26 the default end-to-end head does not use NMS, so iou=0.6 will not change this result; two close boxes usually means the model/data still learned duplicate assignments, and seeing a DFL entry in best.pt alone does not mean the checkpoint is not YOLO26. Please verify the model is actually running the end-to-end head as shown in the end-to-end detection guide, and if your labels are segmentation polygons retrain with yolo26m-seg.pt instead of yolo26m.pt; if it still reproduces, please paste the outputs below and 2-3 raw image/label pairs.

yolo checks

from ultralytics import YOLO

model = YOLO("best.pt")
model.predict("image.jpg", verbose=False)
print(model.predictor.model.end2end)

[Y-T-G]

How many images do you have for training?