TexasInstruments-EdgeAI/YOLO11-Detection

YOLO11 for TI EdgeAI

Real-Time Object Detector with an Enhanced Backbone and C2PSA Attention

---

Overview

YOLO11 is Ultralytics' successor to YOLOv8, released in September 2024. It keeps the overall one-stage, anchor-free detection pipeline but reworks the backbone and neck for more efficient feature extraction: the C2f block used throughout YOLOv8 is replaced by a C3k2 block (a faster variant of the CSP bottleneck that can switch between smaller convolution kernels for efficiency), and a C2PSA (Cross-Stage Partial with Spatial Attention) module is inserted after the backbone's SPPF layer to add lightweight spatial-attention refinement to the extracted features. Together with a refined training pipeline, these changes let YOLO11 reach higher COCO mAP than the equivalent YOLOv8 scale while using noticeably fewer parameters — for example, YOLO11m matches or beats YOLOv8m's accuracy with about 22% fewer parameters.

YOLO11 is offered in five size variants — n, s, m, l, x — spanning a wide accuracy-speed trade-off, from resource-constrained edge devices up to high-throughput deployments. This model is optimized for Texas Instruments MPU (Microprocessor Unit) devices, targeting edge computer vision use cases such as industrial automation, smart cameras, robotics, and IoT vision.

See YOLO26 for the newest Ultralytics generation, with native end-to-end (NMS-free) detection.

---

Model Variants

Model	Params (M)	Input Size	mAP[.5:.95]%	Validated Devices	Config
yolo11n	2.6	640×640	39.5	TDA4VH, TDA4VL	yolo11n_model_config.yaml
yolo11s	9.4	640×640	47.0	TDA4VH, TDA4VL	yolo11s_model_config.yaml
yolo11m	20.1	640×640	51.5	TDA4VH, TDA4VL	yolo11m_model_config.yaml
yolo11l	25.3	640×640	53.4	TDA4VH, TDA4VL	yolo11l_model_config.yaml
yolo11x	56.9	640×640	54.7	TDA4VH, TDA4VL	yolo11x_model_config.yaml

Recommended for edge deployment: yolo11n (best accuracy/compute trade-off, smallest footprint)

---

Quick Start

Prerequisites

pip install onnx>=1.22.0
pip install onnxruntime>=1.23.2
pip install ultralytics

For TI hardware deployment, also set up tidlrunner.

If accessing this model from HuggingFace, clone the repository using the hf CLI:

hf download <REPO_ID> --local-dir <download_location>

Export the Model

# Prepare the default model (yolo11n)
python prepare_model.py

# Prepare a specific model variant
python prepare_model.py --model yolo11s

# Prepare multiple variants in one run
python prepare_model.py --model yolo11n yolo11s yolo11m

# Prepare every supported variant
python prepare_model.py --model all

# List all supported variants and their local download/conversion status
python prepare_model.py --list-models

# Re-run shape fixing on an already-downloaded ONNX
python prepare_model.py --model yolo11n --skip-download

The script automatically:

• Parses the variant's .link file to get the HuggingFace download URL for the .pt checkpoint
• Downloads the .pt model with curl if it isn't already present locally
• Converts the .pt model to ONNX (opset 17) using Ultralytics' model.export()
• Fixes dynamic input dimensions to a static shape (default [1, 3, 640, 640])
• Runs ONNX shape inference and optional onnx-simplifier optimization, then validates the result

Compile and Infer uing edgeai-tidlrunner

Note: Run the commands below from inside the tidlrunner directory (the cloned edgeai-tidlrunner repository), with --config_path pointing to this model's config file.

Compile using edgeai-tidlrunner - on PC

cd /path/to/edgeai-tidlrunner
tidlrunner-cli compile --target_device J784S4 \
  --config_path /path/to/yolo11n_model_config.yaml

Run Inference Benchmark - on device

cd /path/to/edgeai-tidlrunner
tidlrunner-cli infer --target_device J784S4 \
  --config_path /path/to/yolo11n_model_config.yaml

To evaluate accuracy instead, replace infer with evaluate in the command above.

Compile and Infer using edgeai-tidl-tools (Advanced):

Follow the instructions at https://github.com/TexasInstruments/edgeai-tidl-tools

Deploy using edgeai-tidl-tools:

Deplyment can be done using edgeai-tidl-tools. For ONNX models, onnxruntime-tidl with TIDL acceleration can be used. Consult the documentation of edgeai-tidl-tools for more details.

---

Citation

Ultralytics has not published a formal research paper for YOLO11 due to the rapidly evolving nature of the models. If you use the YOLO11 model or any other software from the Ultralytics repository in your work, please cite it using the following format:

@software{yolo11_ultralytics,
  author = {Glenn Jocher and Jing Qiu},
  title = {Ultralytics YOLO11},
  version = {11.0.0},
  year = {2024},
  url = {https://github.com/ultralytics/ultralytics},
  orcid = {0000-0001-5950-6979, 0000-0003-3783-7069},
  license = {AGPL-3.0}
}

---

🔗 Resources

Resource	Link
Source Code	ultralytics/ultralytics
Documentation	YOLO11 Docs
edgeai-tidl-tools	GitHub
edgeai-tidlrunner	GitHub
EdgeAI SDK	Documentation
EdgeAI MPU Overview	GitHub
TI EdgeAI Ecosystem	GitHub

---

Related Models

YOLOv8 Predecessor generation Anchor-free split head

YOLO26 Newest Ultralytics generation NMS-free end-to-end detection

YOLOX Anchor-free YOLO variant Decoupled head design

RTMDet CNN-based alternative Real-time mmdetection model

---

Maintained by: Texas Instruments EdgeAI Team
Last Updated: August 2026