YOLO-NAS-m
Object Detection
INT8
W8A16
FP16
post
YOLO-NAS-m: Object Detection

YOLO-NAS is a next-generation real-time object detection model optimized via Neural Architecture Search (NAS), automating the balance between accuracy and speed for superior performance in complex scenarios. It integrates hybrid quantization-aware architectures with reparameterized blocks and dynamic sparse attention, enhancing small/occluded object detection while reducing computation. Through multi-objective optimization (e.g., latency, parameters, mAP), it discovers efficient structures supporting FP16/INT8 quantization, achieving ~5% higher mAP than YOLOv8 on COCO with 80+ FPS on mobile GPUs. Ideal for autonomous driving and surveillance, it balances edge-device constraints and high precision, offering flexible speed-accuracy tradeoffs.

Source model

  • Input shape: 1x3x640x640
  • Number of parameters: 30.44M
  • Model size: 121.87M
  • Output shape: 1x8400x4, 1x8400x80

The source model can be found here

Performance Reference
Device
Backend
Precision
Inference Time
Accuracy Loss
File Size
Model Optimization

When the user has fine-tuned the source model, the model conversion process must be performed again.

Users can refer to either of the following two methods to complete the model conversion:

  • Using AIMO for model conversion: Click Model Conversion Reference in the Performance Reference section on the right to view the conversion steps.

  • Using Qualcomm QNN for model conversion: Please refer to the Qualcomm QNN Documentation.

Model Inference

The model performance benchmarks and example code provided by Model Farm are all implemented based on the AIDLux AidLite SDK.

For models in .bin format, you can use either of the following two inference engines to run inference on Qualcomm chips:

Inference using AidLite SDK

For more details, please refer to the AidLite Developer Documentation

  • Install AidLite SDK
# install aidlite sdk c++ api
sudo aid-pkg -i aidlite-sdk

# install aidlite sdk python api
python3 -m pip install pyaidlite -i https://mirrors.aidlux.com --trusted-host mirrors.aidlux.com
  • Verify AidLite SDK installation
# aidlite sdk c++ check
python3 -c "import aidlite; print(aidlite.get_library_version())"

# aidlite sdk python check
python3 -c "import aidlite; print(aidlite.get_py_library_version())"

Inference using Qualcomm QNN

Please refer to the Qualcomm QNN Documentation

Inference Example Code

The inference example code is implemented using the AidLite SDK.

  • Click Model & Code to download the model files and the inference code package. The file structure is as follows:
/model_farm_{model_name}_aidlite
    
    |__ models # folder where model files are stored

    |__ python # aidlite python model inference example

    |__ cpp # aidlite cpp model inference example

    |__ README.md
License
Source Model:YOLO-NAS License
Deployable Model:YOLO-NAS License
Performance Reference
Device
Backend
Precision
Inference Time
Accuracy Loss
File Size