Faster Segment Anything: Towards Lightweight SAM for Mobile Applications
Paper
•
2306.14289
•
Published
•
15
MobileSam: Optimized for Mobile Deployment
Faster Segment Anything: Towards lightweight SAM for mobile applications
Transformer based encoder-decoder where prompts specify what to segment in an image thereby allowing segmentation without the need for additional training. The image encoder generates embeddings and the lightweight decoder operates on the embeddings for point and mask based image segmentation.
This model is an implementation of MobileSam found here.
This repository provides scripts to run MobileSam on Qualcomm® devices. More details on model performance across various devices, can be found here.
Model Details
- Model Type: Model_use_case.semantic_segmentation
- Model Stats:
- Model checkpoint: vit_t
- Input resolution: 720p (720x1280)
- Number of parameters (SAMEncoder): 6.95M
- Model size (SAMEncoder) (float): 26.6 MB
- Number of parameters (SAMDecoder): 6.16M
- Model size (SAMDecoder) (float): 23.7 MB
| Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
|---|---|---|---|---|---|---|---|---|
| MobileSAMEncoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 289.017 ms | 4 - 1124 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 208.058 ms | 4 - 1239 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 592.229 ms | 4 - 3349 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 453.257 ms | 12 - 2221 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 172.788 ms | 2 - 6 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 101.539 ms | 12 - 15 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 342.21 ms | 108 - 124 MB | NPU | MobileSam.onnx.zip |
| MobileSAMEncoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 621.611 ms | 4 - 1130 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 101.469 ms | 1 - 1248 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 289.017 ms | 4 - 1124 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 208.058 ms | 4 - 1239 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 173.792 ms | 4 - 8 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 102.152 ms | 12 - 19 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 484.725 ms | 4 - 1772 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 416.048 ms | 0 - 1357 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 175.02 ms | 4 - 9 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 102.099 ms | 12 - 14 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 621.611 ms | 4 - 1130 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 101.469 ms | 1 - 1248 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 126.386 ms | 4 - 1374 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 72.959 ms | 12 - 2876 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 250.191 ms | 74 - 816 MB | NPU | MobileSam.onnx.zip |
| MobileSAMEncoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 96.798 ms | 3 - 1075 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 50.062 ms | 12 - 1177 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 186.071 ms | 117 - 759 MB | NPU | MobileSam.onnx.zip |
| MobileSAMEncoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 78.554 ms | 4 - 1142 MB | NPU | MobileSam.tflite |
| MobileSAMEncoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 40.246 ms | 14 - 872 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 153.31 ms | 78 - 637 MB | NPU | MobileSam.onnx.zip |
| MobileSAMEncoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 100.617 ms | 12 - 12 MB | NPU | MobileSam.dlc |
| MobileSAMEncoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 356.544 ms | 132 - 132 MB | NPU | MobileSam.onnx.zip |
| MobileSAMDecoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 13.139 ms | 0 - 182 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 11.878 ms | 1 - 184 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 10.646 ms | 0 - 202 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 8.46 ms | 4 - 201 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 5.434 ms | 0 - 18 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 4.817 ms | 4 - 6 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | ONNX | 8.374 ms | 0 - 14 MB | NPU | MobileSam.onnx.zip |
| MobileSAMDecoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 6.625 ms | 0 - 186 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 5.832 ms | 1 - 184 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 13.139 ms | 0 - 182 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 11.878 ms | 1 - 184 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 5.436 ms | 0 - 8 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 4.822 ms | 4 - 6 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 8.1 ms | 0 - 179 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 7.261 ms | 0 - 204 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 5.451 ms | 0 - 18 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 4.814 ms | 4 - 6 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 6.625 ms | 0 - 186 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 5.832 ms | 1 - 184 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 3.825 ms | 0 - 212 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 3.304 ms | 4 - 214 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | ONNX | 5.752 ms | 5 - 218 MB | NPU | MobileSam.onnx.zip |
| MobileSAMDecoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 2.899 ms | 0 - 173 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 2.563 ms | 0 - 171 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | ONNX | 4.284 ms | 2 - 180 MB | NPU | MobileSam.onnx.zip |
| MobileSAMDecoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | TFLITE | 2.403 ms | 0 - 174 MB | NPU | MobileSam.tflite |
| MobileSAMDecoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | QNN_DLC | 2.166 ms | 3 - 172 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen5 Mobile | ONNX | 3.507 ms | 0 - 185 MB | NPU | MobileSam.onnx.zip |
| MobileSAMDecoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 5.266 ms | 4 - 4 MB | NPU | MobileSam.dlc |
| MobileSAMDecoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | ONNX | 8.693 ms | 11 - 11 MB | NPU | MobileSam.onnx.zip |
Installation
Install the package via pip:
# NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported.
pip install "qai-hub-models[mobilesam]" 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.mobilesam.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.mobilesam.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.mobilesam.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.mobilesam 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.mobilesam.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.mobilesam.demo -- --eval-mode on-device
Deploying compiled model to Android
The models can be deployed using multiple runtimes:
TensorFlow Lite (
.tfliteexport): This tutorial provides a guide to deploy the .tflite model in an Android application.QNN (
.soexport ): This sample app provides instructions on how to use the.soshared library in an Android application.
View on Qualcomm® AI Hub
Get more details on MobileSam's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of MobileSam 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
- 175