Segment-Anything-Model-2: Optimized for Mobile Deployment
High-quality segmentation in images and videos with real-time performance and minimal user interaction
SAM 2, the successor to Meta's Segment Anything Model (SAM), is a cutting-edge tool designed for comprehensive object segmentation in both images and videos. It excels in handling complex visual data through a unified, promptable model architecture that supports real-time processing and zero-shot generalization.
This model is an implementation of Segment-Anything-Model-2 found here.
This repository provides scripts to run Segment-Anything-Model-2 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: sam2.1_hiera_t
- Input resolution: 720p (720x1280)
- Number of parameters (SAM2Encoder): 33.5M
- Model size (SAM2Encoder) (float): 128 MB
- Number of parameters (SAM2Decoder): 6.22M
- Model size (SAM2Decoder) (float): 23.7 MB
Model | Precision | Device | Chipset | Target Runtime | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit | Target Model |
---|---|---|---|---|---|---|---|---|
SAM2Encoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 609.767 ms | 92 - 296 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 367.288 ms | 85 - 296 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 384.843 ms | 12 - 2403 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Encoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 252.662 ms | 92 - 117 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 212.081 ms | 15 - 86 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Encoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 290.379 ms | 92 - 295 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 609.767 ms | 92 - 296 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 259.454 ms | 92 - 117 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 214.113 ms | 12 - 93 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Encoder | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 380.286 ms | 92 - 307 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 263.816 ms | 92 - 117 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 214.315 ms | 12 - 89 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Encoder | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 290.379 ms | 92 - 295 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 265.703 ms | 92 - 115 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 215.128 ms | 12 - 87 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Encoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 191.967 ms | 92 - 308 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 178.194 ms | 90 - 298 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Encoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 139.549 ms | 12 - 1333 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Encoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 260.348 ms | 800 - 800 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Decoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 17.97 ms | 0 - 54 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 10.167 ms | 0 - 57 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 11.324 ms | 15 - 76 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Decoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 8.133 ms | 0 - 30 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 6.962 ms | 16 - 35 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Decoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 9.556 ms | 0 - 55 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 17.97 ms | 0 - 54 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | TFLITE | 8.173 ms | 0 - 33 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | SA8255 (Proxy) | Qualcomm® SA8255P (Proxy) | QNN_DLC | 6.96 ms | 16 - 39 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Decoder | float | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 11.752 ms | 0 - 52 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | TFLITE | 8.166 ms | 0 - 32 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | SA8650 (Proxy) | Qualcomm® SA8650P (Proxy) | QNN_DLC | 6.987 ms | 16 - 37 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Decoder | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 9.556 ms | 0 - 55 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | TFLITE | 8.209 ms | 0 - 33 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | Samsung Galaxy S23 | Snapdragon® 8 Gen 2 Mobile | QNN_DLC | 6.982 ms | 15 - 38 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Decoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 5.59 ms | 0 - 65 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | TFLITE | 5.651 ms | 0 - 56 MB | NPU | Segment-Anything-Model-2.tflite |
SAM2Decoder | float | Snapdragon 8 Elite QRD | Snapdragon® 8 Elite Mobile | QNN_DLC | 4.238 ms | 8 - 62 MB | NPU | Segment-Anything-Model-2.dlc |
SAM2Decoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 7.652 ms | 16 - 16 MB | NPU | Segment-Anything-Model-2.dlc |
Installation
Install the package via pip:
pip install "qai-hub-models[sam2]"
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.sam2.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.sam2.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.sam2.export
Profiling Results
------------------------------------------------------------
SAM2Encoder
Device : cs_8275 (ANDROID 14)
Runtime : TFLITE
Estimated inference time (ms) : 609.8
Estimated peak memory usage (MB): [92, 296]
Total # Ops : 647
Compute Unit(s) : npu (593 ops) gpu (0 ops) cpu (54 ops)
------------------------------------------------------------
SAM2Decoder
Device : cs_8275 (ANDROID 14)
Runtime : TFLITE
Estimated inference time (ms) : 18.0
Estimated peak memory usage (MB): [0, 54]
Total # Ops : 889
Compute Unit(s) : npu (889 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.sam2 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.
Run demo on a cloud-hosted device
You can also run the demo on-device.
python -m qai_hub_models.models.sam2.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.sam2.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 Segment-Anything-Model-2's performance across various devices here. Explore all available models on Qualcomm® AI Hub
License
- The license for the original implementation of Segment-Anything-Model-2 can be found here.
- The license for the compiled assets for on-device deployment 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
- 22