Hugging Face's logo

qualcomm
/
Track-Anything

PyTorch
LiteRT
android
video-object-tracking
Model card Files
xet
Community
YAML Metadata Warning: The pipeline tag "video-object-tracking" is not in the official list: text-classification, token-classification, table-question-answering, question-answering, zero-shot-classification, translation, summarization, feature-extraction, text-generation, fill-mask, sentence-similarity, text-to-speech, text-to-audio, automatic-speech-recognition, audio-to-audio, audio-classification, audio-text-to-text, voice-activity-detection, depth-estimation, image-classification, object-detection, image-segmentation, text-to-image, image-to-text, image-to-image, image-to-video, unconditional-image-generation, video-classification, reinforcement-learning, robotics, tabular-classification, tabular-regression, tabular-to-text, table-to-text, multiple-choice, text-ranking, text-retrieval, time-series-forecasting, text-to-video, image-text-to-text, image-text-to-image, image-text-to-video, visual-question-answering, document-question-answering, zero-shot-image-classification, graph-ml, mask-generation, zero-shot-object-detection, text-to-3d, image-to-3d, image-feature-extraction, video-text-to-text, keypoint-detection, visual-document-retrieval, any-to-any, video-to-video, other

Track-Anything: Optimized for Mobile Deployment

High-performance interactive tracking and segmentation in videos

Track-Anything is a video based machine learning model to track an object in a video.

This model is an implementation of Track-Anything found here.

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

Model Details

  • Model Type: Model_use_case.video_object_tracking
  • Model Stats:
    • Model checkpoint: xmem
    • Input resolution: 320x568
    • Number of parameters (TrackAnythingEncodeKeyWithShrinkage): 9.72M
    • Model size (TrackAnythingEncodeKeyWithShrinkage) (float): 37.1 MB
    • Number of parameters (TrackAnythingEncodeValue): 23.3M
    • Model size (TrackAnythingEncodeValue) (float): 88.8 MB
    • Number of parameters (TrackAnythingEncodeKeyWithoutShrinkage): 9.71M
    • Model size (TrackAnythingEncodeKeyWithoutShrinkage) (float): 37.1 MB
    • Number of parameters (TrackAnythingSegment): 30.1M
    • Model size (TrackAnythingSegment) (float): 115 MB
Model Precision Device Chipset Target Runtime Inference Time (ms) Peak Memory Range (MB) Primary Compute Unit Target Model
TrackAnythingEncodeKeyWithShrinkage float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 31.529 ms 3 - 153 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 12.005 ms 3 - 180 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 5.887 ms 3 - 6 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 5.511 ms 0 - 21 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithShrinkage float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 9.687 ms 3 - 154 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float SA7255P ADP Qualcomm® SA7255P TFLITE 31.529 ms 3 - 153 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 5.91 ms 3 - 6 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float SA8295P ADP Qualcomm® SA8295P TFLITE 10.079 ms 3 - 152 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 5.91 ms 3 - 6 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float SA8775P ADP Qualcomm® SA8775P TFLITE 9.687 ms 3 - 154 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 4.025 ms 0 - 189 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 3.902 ms 0 - 145 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithShrinkage float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile TFLITE 3.259 ms 0 - 155 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 3.154 ms 1 - 115 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithShrinkage float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile TFLITE 2.658 ms 0 - 156 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithShrinkage float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile ONNX 2.557 ms 0 - 116 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithShrinkage float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 5.551 ms 19 - 19 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeValue float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 52.329 ms 3 - 201 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 21.505 ms 3 - 257 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 12.234 ms 3 - 7 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 15.677 ms 0 - 44 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeValue float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 17.516 ms 3 - 201 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float SA7255P ADP Qualcomm® SA7255P TFLITE 52.329 ms 3 - 201 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 12.258 ms 4 - 7 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float SA8295P ADP Qualcomm® SA8295P TFLITE 19.451 ms 3 - 198 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 12.256 ms 3 - 5 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float SA8775P ADP Qualcomm® SA8775P TFLITE 17.516 ms 3 - 201 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 8.659 ms 0 - 261 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 9.994 ms 12 - 165 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeValue float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile TFLITE 7.196 ms 3 - 174 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 8.451 ms 7 - 125 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeValue float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile TFLITE 5.718 ms 2 - 172 MB NPU Track-Anything.tflite
TrackAnythingEncodeValue float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile ONNX 6.479 ms 12 - 132 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeValue float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 12.364 ms 38 - 38 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithoutShrinkage float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 33.073 ms 20 - 177 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 13.522 ms 20 - 214 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 7.203 ms 20 - 23 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 7.04 ms 48 - 68 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithoutShrinkage float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 45.236 ms 20 - 178 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float SA7255P ADP Qualcomm® SA7255P TFLITE 33.073 ms 20 - 177 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 7.466 ms 20 - 22 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float SA8295P ADP Qualcomm® SA8295P TFLITE 12.139 ms 20 - 182 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 7.309 ms 17 - 19 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float SA8775P ADP Qualcomm® SA8775P TFLITE 45.236 ms 20 - 178 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 5.214 ms 19 - 230 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 5.27 ms 63 - 208 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithoutShrinkage float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile TFLITE 4.417 ms 18 - 178 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 4.629 ms 61 - 207 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithoutShrinkage float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile TFLITE 3.689 ms 0 - 164 MB NPU Track-Anything.tflite
TrackAnythingEncodeKeyWithoutShrinkage float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile ONNX 3.763 ms 0 - 154 MB NPU Track-Anything.onnx.zip
TrackAnythingEncodeKeyWithoutShrinkage float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 6.909 ms 66 - 66 MB NPU Track-Anything.onnx.zip
TrackAnythingSegment float QCS8275 (Proxy) Qualcomm® QCS8275 (Proxy) TFLITE 103.906 ms 0 - 171 MB NPU Track-Anything.tflite
TrackAnythingSegment float QCS8450 (Proxy) Qualcomm® QCS8450 (Proxy) TFLITE 48.49 ms 2 - 254 MB NPU Track-Anything.tflite
TrackAnythingSegment float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) TFLITE 24.19 ms 2 - 5 MB NPU Track-Anything.tflite
TrackAnythingSegment float QCS8550 (Proxy) Qualcomm® QCS8550 (Proxy) ONNX 23.596 ms 22 - 24 MB NPU Track-Anything.onnx.zip
TrackAnythingSegment float QCS9075 (Proxy) Qualcomm® QCS9075 (Proxy) TFLITE 33.853 ms 2 - 173 MB NPU Track-Anything.tflite
TrackAnythingSegment float SA7255P ADP Qualcomm® SA7255P TFLITE 103.906 ms 0 - 171 MB NPU Track-Anything.tflite
TrackAnythingSegment float SA8255 (Proxy) Qualcomm® SA8255P (Proxy) TFLITE 24.186 ms 2 - 5 MB NPU Track-Anything.tflite
TrackAnythingSegment float SA8295P ADP Qualcomm® SA8295P TFLITE 37.557 ms 2 - 179 MB NPU Track-Anything.tflite
TrackAnythingSegment float SA8650 (Proxy) Qualcomm® SA8650P (Proxy) TFLITE 24.122 ms 1 - 3 MB NPU Track-Anything.tflite
TrackAnythingSegment float SA8775P ADP Qualcomm® SA8775P TFLITE 33.853 ms 2 - 173 MB NPU Track-Anything.tflite
TrackAnythingSegment float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile TFLITE 17.652 ms 1 - 255 MB NPU Track-Anything.tflite
TrackAnythingSegment float Samsung Galaxy S24 Snapdragon® 8 Gen 3 Mobile ONNX 17.523 ms 25 - 211 MB NPU Track-Anything.onnx.zip
TrackAnythingSegment float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile TFLITE 13.485 ms 1 - 180 MB NPU Track-Anything.tflite
TrackAnythingSegment float Samsung Galaxy S25 Snapdragon® 8 Elite For Galaxy Mobile ONNX 14.038 ms 5 - 166 MB NPU Track-Anything.onnx.zip
TrackAnythingSegment float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile TFLITE 10.799 ms 2 - 182 MB NPU Track-Anything.tflite
TrackAnythingSegment float Snapdragon 8 Elite Gen 5 QRD Snapdragon® 8 Elite Gen5 Mobile ONNX 10.804 ms 25 - 187 MB NPU Track-Anything.onnx.zip
TrackAnythingSegment float Snapdragon X Elite CRD Snapdragon® X Elite ONNX 23.754 ms 37 - 37 MB NPU Track-Anything.onnx.zip

Installation

Install the package via pip:

# NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
pip install "qai-hub-models[track-anything]" git+https://github.com/ChaoningZhang/MobileSAM@34bbbfd --use-pep517

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.track_anything.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.track_anything.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.track_anything.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.track_anything 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.

Run demo on a cloud-hosted device

You can also run the demo on-device.

python -m qai_hub_models.models.track_anything.demo --eval-mode on-device

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.track_anything.demo -- --eval-mode on-device

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 Track-Anything's performance across various devices here. Explore all available models on Qualcomm® AI Hub

License

  • The license for the original implementation of Track-Anything can be found here.

References

Community

Downloads last month
72
Inference Providers NEW
This model isn't deployed by any Inference Provider. 🙋 Ask for provider support

Paper for qualcomm/Track-Anything