BEVFusion: Optimized for Mobile Deployment

Construct a bird’s eye view from sensors mounted on a vehicle

BeVFusion is a machine learning model for generating a birds eye view represenation from the sensors(cameras) mounted on a vehicle.

This model is an implementation of BEVFusion found here.

This repository provides scripts to run BEVFusion on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

Model Type: Model_use_case.driver_assistance
Model Stats:
- Model checkpoint: camera-only-det.pth
- Input resolution: 1 x 6 x 3 x 256 x 704
- Number of parameters: 44M
- Model size: 171 MB

Model	Precision	Device	Chipset	Target Runtime	Inference Time (ms)	Peak Memory Range (MB)	Primary Compute Unit	Target Model
BEVFusionEncoder1	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	PRECOMPILED_QNN_ONNX	847.695 ms	0 - 101 MB	NPU	Use Export Script
BEVFusionEncoder1	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	2332.065 ms	9 - 18 MB	NPU	Use Export Script
BEVFusionEncoder1	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	529.22 ms	12 - 31 MB	NPU	Use Export Script
BEVFusionEncoder1	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	628.085 ms	31 - 49 MB	NPU	Use Export Script
BEVFusionEncoder1	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_CONTEXT_BINARY	418.089 ms	12 - 29 MB	NPU	Use Export Script
BEVFusionEncoder1	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	PRECOMPILED_QNN_ONNX	505.178 ms	21 - 35 MB	NPU	Use Export Script
BEVFusionEncoder1	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_CONTEXT_BINARY	371.182 ms	14 - 25 MB	NPU	Use Export Script
BEVFusionEncoder1	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	PRECOMPILED_QNN_ONNX	429.551 ms	50 - 60 MB	NPU	Use Export Script
BEVFusionEncoder1	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	686.665 ms	12 - 12 MB	NPU	Use Export Script
BEVFusionEncoder1	float	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	817.418 ms	97 - 97 MB	NPU	Use Export Script
BEVFusionEncoder2	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	PRECOMPILED_QNN_ONNX	3405.452 ms	586 - 588 MB	NPU	Use Export Script
BEVFusionEncoder2	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	3461.725 ms	16 - 26 MB	NPU	Use Export Script
BEVFusionEncoder2	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	2792.723 ms	17 - 31 MB	NPU	Use Export Script
BEVFusionEncoder2	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	2579.968 ms	493 - 511 MB	NPU	Use Export Script
BEVFusionEncoder2	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_CONTEXT_BINARY	2431.991 ms	17 - 33 MB	NPU	Use Export Script
BEVFusionEncoder2	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	PRECOMPILED_QNN_ONNX	2393.946 ms	433 - 447 MB	NPU	Use Export Script
BEVFusionEncoder2	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_CONTEXT_BINARY	2164.259 ms	17 - 27 MB	NPU	Use Export Script
BEVFusionEncoder2	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	PRECOMPILED_QNN_ONNX	2112.13 ms	286 - 295 MB	NPU	Use Export Script
BEVFusionEncoder2	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	3489.403 ms	17 - 17 MB	NPU	Use Export Script
BEVFusionEncoder2	float	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	3243.578 ms	1058 - 1058 MB	NPU	Use Export Script
BEVFusionEncoder3	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	PRECOMPILED_QNN_ONNX	552.451 ms	608 - 611 MB	NPU	Use Export Script
BEVFusionEncoder3	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	717.337 ms	608 - 618 MB	NPU	Use Export Script
BEVFusionEncoder3	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	589.592 ms	608 - 628 MB	NPU	Use Export Script
BEVFusionEncoder3	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	455.062 ms	587 - 607 MB	NPU	Use Export Script
BEVFusionEncoder3	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_CONTEXT_BINARY	510.774 ms	609 - 625 MB	NPU	Use Export Script
BEVFusionEncoder3	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	PRECOMPILED_QNN_ONNX	377.724 ms	575 - 594 MB	NPU	Use Export Script
BEVFusionEncoder3	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_CONTEXT_BINARY	369.834 ms	609 - 623 MB	NPU	Use Export Script
BEVFusionEncoder3	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	PRECOMPILED_QNN_ONNX	332.66 ms	608 - 622 MB	NPU	Use Export Script
BEVFusionEncoder3	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	698.904 ms	610 - 610 MB	NPU	Use Export Script
BEVFusionEncoder3	float	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	514.14 ms	610 - 610 MB	NPU	Use Export Script
BEVFusionEncoder4	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	PRECOMPILED_QNN_ONNX	12.947 ms	23 - 28 MB	NPU	Use Export Script
BEVFusionEncoder4	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	14.703 ms	18 - 27 MB	NPU	Use Export Script
BEVFusionEncoder4	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	8.786 ms	18 - 37 MB	NPU	Use Export Script
BEVFusionEncoder4	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	9.383 ms	32 - 50 MB	NPU	Use Export Script
BEVFusionEncoder4	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_CONTEXT_BINARY	7.875 ms	2 - 18 MB	NPU	Use Export Script
BEVFusionEncoder4	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	PRECOMPILED_QNN_ONNX	8.115 ms	12 - 26 MB	NPU	Use Export Script
BEVFusionEncoder4	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_CONTEXT_BINARY	6.791 ms	18 - 30 MB	NPU	Use Export Script
BEVFusionEncoder4	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	PRECOMPILED_QNN_ONNX	7.116 ms	31 - 41 MB	NPU	Use Export Script
BEVFusionEncoder4	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	11.783 ms	19 - 19 MB	NPU	Use Export Script
BEVFusionEncoder4	float	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	11.941 ms	19 - 19 MB	NPU	Use Export Script
BEVFusionDecoder	float	QCS8550 (Proxy)	Qualcomm® QCS8550 (Proxy)	PRECOMPILED_QNN_ONNX	13.008 ms	4 - 32 MB	NPU	Use Export Script
BEVFusionDecoder	float	QCS9075 (Proxy)	Qualcomm® QCS9075 (Proxy)	QNN_CONTEXT_BINARY	107.319 ms	1 - 11 MB	NPU	Use Export Script
BEVFusionDecoder	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	QNN_CONTEXT_BINARY	9.698 ms	5 - 25 MB	NPU	Use Export Script
BEVFusionDecoder	float	Samsung Galaxy S24	Snapdragon® 8 Gen 3 Mobile	PRECOMPILED_QNN_ONNX	9.93 ms	16 - 35 MB	NPU	Use Export Script
BEVFusionDecoder	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	QNN_CONTEXT_BINARY	7.689 ms	5 - 22 MB	NPU	Use Export Script
BEVFusionDecoder	float	Samsung Galaxy S25	Snapdragon® 8 Elite For Galaxy Mobile	PRECOMPILED_QNN_ONNX	7.9 ms	11 - 25 MB	NPU	Use Export Script
BEVFusionDecoder	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	QNN_CONTEXT_BINARY	5.76 ms	5 - 16 MB	NPU	Use Export Script
BEVFusionDecoder	float	Snapdragon 8 Elite Gen 5 QRD	Snapdragon® 8 Elite Gen5 Mobile	PRECOMPILED_QNN_ONNX	6.22 ms	17 - 27 MB	NPU	Use Export Script
BEVFusionDecoder	float	Snapdragon X Elite CRD	Snapdragon® X Elite	QNN_CONTEXT_BINARY	13.08 ms	5 - 5 MB	NPU	Use Export Script
BEVFusionDecoder	float	Snapdragon X Elite CRD	Snapdragon® X Elite	PRECOMPILED_QNN_ONNX	13.249 ms	24 - 24 MB	NPU	Use Export Script

Installation

Install the package via pip:

# NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
pip install "qai-hub-models[bevfusion-det]"

Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device

Sign-in to Qualcomm® AI Hub Workbench with your Qualcomm® ID. Once signed in navigate to Account -> Settings -> API Token.

With this API token, you can configure your client to run models on the cloud hosted devices.

qai-hub configure --api_token API_TOKEN

Navigate to docs for more information.

Demo off target

The package contains a simple end-to-end demo that downloads pre-trained weights and runs this model on a sample input.

python -m qai_hub_models.models.bevfusion_det.demo

The above demo runs a reference implementation of pre-processing, model inference, and post processing.

NOTE: If you want running in a Jupyter Notebook or Google Colab like environment, please add the following to your cell (instead of the above).

%run -m qai_hub_models.models.bevfusion_det.demo

Run model on a cloud-hosted device

In addition to the demo, you can also run the model on a cloud-hosted Qualcomm® device. This script does the following:

Performance check on-device on a cloud-hosted device
Downloads compiled assets that can be deployed on-device for Android.
Accuracy check between PyTorch and on-device outputs.

python -m qai_hub_models.models.bevfusion_det.export

How does this work?

This export script leverages Qualcomm® AI Hub to optimize, validate, and deploy this model on-device. Lets go through each step below in detail:

Step 1: Compile model for on-device deployment

To compile a PyTorch model for on-device deployment, we first trace the model in memory using the jit.trace and then call the submit_compile_job API.

import torch

import qai_hub as hub
from qai_hub_models.models.bevfusion_det import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S25")

# Trace model
input_shape = torch_model.get_input_spec()
sample_inputs = torch_model.sample_inputs()

pt_model = torch.jit.trace(torch_model, [torch.tensor(data[0]) for _, data in sample_inputs.items()])

# Compile model on a specific device
compile_job = hub.submit_compile_job(
    model=pt_model,
    device=device,
    input_specs=torch_model.get_input_spec(),
)

# Get target model to run on-device
target_model = compile_job.get_target_model()

Step 2: Performance profiling on cloud-hosted device

After compiling models from step 1. Models can be profiled model on-device using the target_model. Note that this scripts runs the model on a device automatically provisioned in the cloud. Once the job is submitted, you can navigate to a provided job URL to view a variety of on-device performance metrics.

profile_job = hub.submit_profile_job(
    model=target_model,
    device=device,
)

Step 3: Verify on-device accuracy

To verify the accuracy of the model on-device, you can run on-device inference on sample input data on the same cloud hosted device.

input_data = torch_model.sample_inputs()
inference_job = hub.submit_inference_job(
    model=target_model,
    device=device,
    inputs=input_data,
)
    on_device_output = inference_job.download_output_data()

With the output of the model, you can compute like PSNR, relative errors or spot check the output with expected output.

Note: This on-device profiling and inference requires access to Qualcomm® AI Hub Workbench. Sign up for access.

Deploying compiled model to Android

The models can be deployed using multiple runtimes:

TensorFlow Lite (.tflite export): This tutorial provides a guide to deploy the .tflite model in an Android application.
QNN (.so export ): This sample app provides instructions on how to use the .so shared library in an Android application.

View on Qualcomm® AI Hub

Get more details on BEVFusion's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

The license for the original implementation of BEVFusion can be found here.

References

Community

Join our AI Hub Slack community to collaborate, post questions and learn more about on-device AI.
For questions or feedback please reach out to us.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

Other

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Paper for qualcomm/BEVFusion

BEVFusion: Multi-Task Multi-Sensor Fusion with Unified Bird's-Eye View Representation

Paper • 2205.13542 • Published May 26, 2022