3D-Deep-BOX: Optimized for Mobile Deployment

Real-time 3D object detection

3D Deep Box is a machine learning model that predicts 3D bounding boxes and classes of objects in an image.

This model is an implementation of 3D-Deep-BOX found here.

This repository provides scripts to run 3D-Deep-BOX on Qualcomm® devices. More details on model performance across various devices, can be found here.

Model Details

  • Model Type: Model_use_case.object_detection
  • Model Stats:
    • Model checkpoint: YOLOv3-tiny
    • Input resolution(YOLO): 224x640
    • Input resolution(VGG): 224x224
    • Number of parameters (Yolo2DDetection): 9.78M
    • Model size (Yolo2DDetection) (float): 37.3 MB
    • Number of parameters (VGG3DDetection): 46.1M
    • Model size (VGG3DDetection) (float): 176 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
Yolo2DDetection float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 38.346 ms 0 - 29 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 7.463 ms 2 - 39 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 21.123 ms 0 - 59 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 3.298 ms 2 - 39 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 20.628 ms 0 - 151 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 2.384 ms 1 - 41 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 22.077 ms 0 - 30 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 3.106 ms 2 - 37 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float SA7255P ADP Qualcomm® SA7255P TFLITE 38.346 ms 0 - 29 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float SA7255P ADP Qualcomm® SA7255P QNN_DLC 7.463 ms 2 - 39 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 20.793 ms 0 - 145 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 2.374 ms 2 - 16 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float SA8295P ADP Qualcomm® SA8295P TFLITE 23.372 ms 0 - 33 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float SA8295P ADP Qualcomm® SA8295P QNN_DLC 3.49 ms 2 - 34 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 20.807 ms 0 - 149 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 2.384 ms 0 - 38 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float SA8775P ADP Qualcomm® SA8775P TFLITE 22.077 ms 0 - 30 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float SA8775P ADP Qualcomm® SA8775P QNN_DLC 3.106 ms 2 - 37 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 20.904 ms 0 - 145 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 2.377 ms 2 - 15 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 2.638 ms 0 - 69 MB NPU 3D-Deep-BOX.onnx
Yolo2DDetection float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 15.688 ms 0 - 58 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 1.767 ms 2 - 48 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 1.881 ms 1 - 46 MB NPU 3D-Deep-BOX.onnx
Yolo2DDetection float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 16.652 ms 0 - 34 MB NPU 3D-Deep-BOX.tflite
Yolo2DDetection float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 1.824 ms 2 - 40 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 2.071 ms 1 - 37 MB NPU 3D-Deep-BOX.onnx
Yolo2DDetection float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 2.604 ms 40 - 40 MB NPU 3D-Deep-BOX.dlc
Yolo2DDetection float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 2.715 ms 18 - 18 MB NPU 3D-Deep-BOX.onnx
VGG3DDetection float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 32.845 ms 0 - 74 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) QNN_DLC 32.765 ms 0 - 77 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 8.292 ms 0 - 127 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) QNN_DLC 9.623 ms 1 - 84 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 4.73 ms 0 - 740 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) QNN_DLC 4.783 ms 0 - 461 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 8.92 ms 0 - 74 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) QNN_DLC 8.895 ms 1 - 77 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float SA7255P ADP Qualcomm® SA7255P TFLITE 32.845 ms 0 - 74 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float SA7255P ADP Qualcomm® SA7255P QNN_DLC 32.765 ms 0 - 77 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 4.739 ms 0 - 747 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) QNN_DLC 4.778 ms 0 - 459 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float SA8295P ADP Qualcomm® SA8295P TFLITE 9.747 ms 0 - 79 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float SA8295P ADP Qualcomm® SA8295P QNN_DLC 9.728 ms 1 - 81 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 4.745 ms 0 - 705 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) QNN_DLC 4.775 ms 0 - 459 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float SA8775P ADP Qualcomm® SA8775P TFLITE 8.92 ms 0 - 74 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float SA8775P ADP Qualcomm® SA8775P QNN_DLC 8.895 ms 1 - 77 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile TFLITE 4.736 ms 0 - 741 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile QNN_DLC 4.777 ms 0 - 472 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float Samsung Galaxy S23 Snapdragon® 8 Gen 2 Mobile ONNX 5.049 ms 0 - 469 MB NPU 3D-Deep-BOX.onnx
VGG3DDetection float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 3.555 ms 0 - 123 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile QNN_DLC 3.659 ms 1 - 85 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 3.799 ms 0 - 82 MB NPU 3D-Deep-BOX.onnx
VGG3DDetection float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile TFLITE 3.447 ms 0 - 79 MB NPU 3D-Deep-BOX.tflite
VGG3DDetection float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile QNN_DLC 3.494 ms 1 - 80 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float Snapdragon 8 Elite QRD Snapdragon® 8 Elite Mobile ONNX 3.712 ms 1 - 79 MB NPU 3D-Deep-BOX.onnx
VGG3DDetection float Snapdragon X Elite CRD Snapdragon® X Elite QNN_DLC 4.877 ms 449 - 449 MB NPU 3D-Deep-BOX.dlc
VGG3DDetection float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 4.971 ms 89 - 89 MB NPU 3D-Deep-BOX.onnx

Installation

Install the package via pip:

pip install "qai-hub-models[deepbox]"

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

Sign-in to Qualcomm® AI Hub 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.deepbox.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.deepbox.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.deepbox.export
Profiling Results
------------------------------------------------------------
Yolo2DDetection
Device                          : cs_8275 (ANDROID 14)                 
Runtime                         : TFLITE                               
Estimated inference time (ms)   : 38.3                                 
Estimated peak memory usage (MB): [0, 29]                              
Total # Ops                     : 129                                  
Compute Unit(s)                 : npu (129 ops) gpu (0 ops) cpu (0 ops)

------------------------------------------------------------
VGG3DDetection
Device                          : cs_8275 (ANDROID 14)                
Runtime                         : TFLITE                              
Estimated inference time (ms)   : 32.8                                
Estimated peak memory usage (MB): [0, 74]                             
Total # Ops                     : 40                                  
Compute Unit(s)                 : npu (40 ops) gpu (0 ops) cpu (0 ops)

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.deepbox import Model

# Load the model
torch_model = Model.from_pretrained()

# Device
device = hub.Device("Samsung Galaxy S24")

# 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. 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 3D-Deep-BOX's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of 3D-Deep-BOX can be found here.
  • The license for the compiled assets for on-device deployment can be found here

References

Community

Downloads last month
86
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support