Hugging Face's logo

qualcomm
/
3D-Deep-BOX

Object Detection
PyTorch
LiteRT
ONNX
real_time
android
Model card Files Community

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: Object detection
  • Model Stats:
    • Model checkpoint: YOLOv3-tiny
    • Input resolution(YOLO): 224x640
    • Number of parameters(YOLO): 8.85M
    • Model size(YOLO): 37.3 MB
    • Input resolution(VGG): 224x224
    • Number of parameters(VGG): 144M
    • Model size(VGG): 175.9 MB
Model Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Precision Primary Compute Unit Target Model
Yolo2DDetection Samsung Galaxy S23 Snapdragon® 8 Gen 2 TFLITE 22.177 ms 0 - 138 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection Samsung Galaxy S23 Snapdragon® 8 Gen 2 QNN 2.576 ms 0 - 21 MB FP16 NPU 3D-Deep-BOX.so
Yolo2DDetection Samsung Galaxy S23 Snapdragon® 8 Gen 2 ONNX 4.974 ms 0 - 56 MB FP16 NPU 3D-Deep-BOX.onnx
Yolo2DDetection Samsung Galaxy S24 Snapdragon® 8 Gen 3 TFLITE 16.447 ms 0 - 60 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection Samsung Galaxy S24 Snapdragon® 8 Gen 3 QNN 1.817 ms 0 - 23 MB FP16 NPU 3D-Deep-BOX.so
Yolo2DDetection Samsung Galaxy S24 Snapdragon® 8 Gen 3 ONNX 3.401 ms 0 - 27 MB FP16 NPU 3D-Deep-BOX.onnx
Yolo2DDetection Snapdragon 8 Elite QRD Snapdragon® 8 Elite TFLITE 17.458 ms 0 - 32 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection Snapdragon 8 Elite QRD Snapdragon® 8 Elite QNN 1.657 ms 2 - 19 MB FP16 NPU Use Export Script
Yolo2DDetection Snapdragon 8 Elite QRD Snapdragon® 8 Elite ONNX 4.22 ms 1 - 17 MB FP16 NPU 3D-Deep-BOX.onnx
Yolo2DDetection SA7255P ADP SA7255P TFLITE 68.127 ms 0 - 30 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection SA7255P ADP SA7255P QNN 34.633 ms 2 - 11 MB FP16 NPU Use Export Script
Yolo2DDetection SA8255 (Proxy) SA8255P Proxy TFLITE 22.33 ms 0 - 139 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection SA8255 (Proxy) SA8255P Proxy QNN 2.44 ms 2 - 4 MB FP16 NPU Use Export Script
Yolo2DDetection SA8295P ADP SA8295P TFLITE 23.666 ms 0 - 31 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection SA8295P ADP SA8295P QNN 3.495 ms 0 - 18 MB FP16 NPU Use Export Script
Yolo2DDetection SA8650 (Proxy) SA8650P Proxy TFLITE 22.442 ms 0 - 135 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection SA8650 (Proxy) SA8650P Proxy QNN 2.437 ms 2 - 11 MB FP16 NPU Use Export Script
Yolo2DDetection SA8775P ADP SA8775P TFLITE 28.155 ms 0 - 30 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection SA8775P ADP SA8775P QNN 3.859 ms 2 - 12 MB FP16 NPU Use Export Script
Yolo2DDetection QCS8275 (Proxy) QCS8275 Proxy TFLITE 68.127 ms 0 - 30 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection QCS8275 (Proxy) QCS8275 Proxy QNN 34.633 ms 2 - 11 MB FP16 NPU Use Export Script
Yolo2DDetection QCS8550 (Proxy) QCS8550 Proxy TFLITE 22.134 ms 0 - 138 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection QCS8550 (Proxy) QCS8550 Proxy QNN 2.435 ms 2 - 4 MB FP16 NPU Use Export Script
Yolo2DDetection QCS9075 (Proxy) QCS9075 Proxy TFLITE 28.155 ms 0 - 30 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection QCS9075 (Proxy) QCS9075 Proxy QNN 3.859 ms 2 - 12 MB FP16 NPU Use Export Script
Yolo2DDetection QCS8450 (Proxy) QCS8450 Proxy TFLITE 22.338 ms 0 - 59 MB FP16 NPU 3D-Deep-BOX.tflite
Yolo2DDetection QCS8450 (Proxy) QCS8450 Proxy QNN 3.266 ms 2 - 25 MB FP16 NPU Use Export Script
Yolo2DDetection Snapdragon X Elite CRD Snapdragon® X Elite QNN 2.622 ms 2 - 2 MB FP16 NPU Use Export Script
Yolo2DDetection Snapdragon X Elite CRD Snapdragon® X Elite ONNX 5.6 ms 3 - 3 MB FP16 NPU 3D-Deep-BOX.onnx
VGG3DDetection Samsung Galaxy S23 Snapdragon® 8 Gen 2 TFLITE 4.736 ms 0 - 668 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection Samsung Galaxy S23 Snapdragon® 8 Gen 2 QNN 4.88 ms 0 - 429 MB FP16 NPU 3D-Deep-BOX.so
VGG3DDetection Samsung Galaxy S23 Snapdragon® 8 Gen 2 ONNX 5.097 ms 0 - 413 MB FP16 NPU 3D-Deep-BOX.onnx
VGG3DDetection Samsung Galaxy S24 Snapdragon® 8 Gen 3 TFLITE 3.547 ms 0 - 125 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection Samsung Galaxy S24 Snapdragon® 8 Gen 3 QNN 3.691 ms 1 - 83 MB FP16 NPU 3D-Deep-BOX.so
VGG3DDetection Samsung Galaxy S24 Snapdragon® 8 Gen 3 ONNX 3.935 ms 0 - 86 MB FP16 NPU 3D-Deep-BOX.onnx
VGG3DDetection Snapdragon 8 Elite QRD Snapdragon® 8 Elite TFLITE 2.947 ms 0 - 79 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection Snapdragon 8 Elite QRD Snapdragon® 8 Elite QNN 3.428 ms 1 - 76 MB FP16 NPU Use Export Script
VGG3DDetection Snapdragon 8 Elite QRD Snapdragon® 8 Elite ONNX 3.799 ms 1 - 79 MB FP16 NPU 3D-Deep-BOX.onnx
VGG3DDetection SA7255P ADP SA7255P TFLITE 257.878 ms 0 - 73 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection SA7255P ADP SA7255P QNN 257.883 ms 1 - 10 MB FP16 NPU Use Export Script
VGG3DDetection SA8255 (Proxy) SA8255P Proxy TFLITE 4.727 ms 0 - 674 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection SA8255 (Proxy) SA8255P Proxy QNN 4.778 ms 1 - 3 MB FP16 NPU Use Export Script
VGG3DDetection SA8295P ADP SA8295P TFLITE 9.755 ms 0 - 76 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection SA8295P ADP SA8295P QNN 9.925 ms 1 - 18 MB FP16 NPU Use Export Script
VGG3DDetection SA8650 (Proxy) SA8650P Proxy TFLITE 4.732 ms 0 - 672 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection SA8650 (Proxy) SA8650P Proxy QNN 4.776 ms 1 - 3 MB FP16 NPU Use Export Script
VGG3DDetection SA8775P ADP SA8775P TFLITE 10.792 ms 0 - 73 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection SA8775P ADP SA8775P QNN 10.698 ms 1 - 10 MB FP16 NPU Use Export Script
VGG3DDetection QCS8275 (Proxy) QCS8275 Proxy TFLITE 257.878 ms 0 - 73 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection QCS8275 (Proxy) QCS8275 Proxy QNN 257.883 ms 1 - 10 MB FP16 NPU Use Export Script
VGG3DDetection QCS8550 (Proxy) QCS8550 Proxy TFLITE 4.735 ms 0 - 663 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection QCS8550 (Proxy) QCS8550 Proxy QNN 4.773 ms 1 - 3 MB FP16 NPU Use Export Script
VGG3DDetection QCS9075 (Proxy) QCS9075 Proxy TFLITE 10.792 ms 0 - 73 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection QCS9075 (Proxy) QCS9075 Proxy QNN 10.698 ms 1 - 10 MB FP16 NPU Use Export Script
VGG3DDetection QCS8450 (Proxy) QCS8450 Proxy TFLITE 8.419 ms 0 - 124 MB FP16 NPU 3D-Deep-BOX.tflite
VGG3DDetection QCS8450 (Proxy) QCS8450 Proxy QNN 8.67 ms 1 - 81 MB FP16 NPU Use Export Script
VGG3DDetection Snapdragon X Elite CRD Snapdragon® X Elite QNN 5.03 ms 1 - 1 MB FP16 NPU Use Export Script
VGG3DDetection Snapdragon X Elite CRD Snapdragon® X Elite ONNX 4.951 ms 88 - 88 MB FP16 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                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 22.2                   
Estimated peak memory usage (MB): [0, 138]               
Total # Ops                     : 129                    
Compute Unit(s)                 : NPU (129 ops)          

------------------------------------------------------------
VGG3DDetection
Device                          : Samsung Galaxy S23 (13)
Runtime                         : TFLITE                 
Estimated inference time (ms)   : 4.7                    
Estimated peak memory usage (MB): [0, 668]               
Total # Ops                     : 40                     
Compute Unit(s)                 : NPU (40 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

-

Downloads are not tracked for this model. How to track
Inference Providers NEW
Object Detection
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support