README.md

---
library_name: pytorch
license: other
tags:
- android
pipeline_tag: image-segmentation

---

![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/maskrcnn/web-assets/model_demo.png)

# MaskRCNN: Optimized for Qualcomm Devices

Mask R-CNN is a machine learning model that extends Faster R-CNN to perform instance segmentation by detecting objects in an image while simultaneously generating a high-quality segmentation mask for each instance. It adds a branch for predicting segmentation masks in parallel with the existing branch for bounding box recognition.

This is based on the implementation of MaskRCNN found [here](https://github.com/pytorch/vision).
This repository contains pre-exported model files optimized for Qualcomm® devices. You can use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/maskrcnn) library to export with custom configurations. More details on model performance across various devices, can be found [here](#performance-summary).

Qualcomm AI Hub Models uses [Qualcomm AI Hub Workbench](https://workbench.aihub.qualcomm.com) to compile, profile, and evaluate this model. [Sign up](https://myaccount.qualcomm.com/signup) to run these models on a hosted Qualcomm® device.

## Getting Started
There are two ways to deploy this model on your device:

### Option 1: Download Pre-Exported Models

Below are pre-exported model assets ready for deployment.

| Runtime | Precision | Chipset | SDK Versions | Download |
|---|---|---|---|---|
| QNN_DLC | float | Universal | QAIRT 2.45 | [Download](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/maskrcnn/releases/v0.53.1/maskrcnn-qnn_dlc-float.zip)

For more device-specific assets and performance metrics, visit **[MaskRCNN on Qualcomm® AI Hub](https://aihub.qualcomm.com/models/maskrcnn)**.


### Option 2: Export with Custom Configurations

Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/maskrcnn) Python library to compile and export the model with your own:
- Custom weights (e.g., fine-tuned checkpoints)
- Custom input shapes
- Target device and runtime configurations

This option is ideal if you need to customize the model beyond the default configuration provided here.

See our repository for [MaskRCNN on GitHub](https://github.com/qualcomm/ai-hub-models/blob/main/src/qai_hub_models/models/maskrcnn) for usage instructions.

## Model Details

**Model Type:** Model_use_case.semantic_segmentation

**Model Stats:**
- Model checkpoint: Mask R-CNN ResNet-50 FPN V2
- Input resolution: 800x800
- Number of output classes: 91
- Number of parameters: 46.4M
- Model size (float): 177 MB

## Performance Summary
| Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit
|---|---|---|---|---|---|---
| proposal_generator | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 44.202 ms | 7 - 1769 MB | NPU
| proposal_generator | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 53.244 ms | 0 - 1510 MB | NPU
| proposal_generator | QNN_DLC | float | Snapdragon® X2 Elite | 43.187 ms | 7 - 7 MB | NPU
| proposal_generator | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 69.276 ms | 7 - 2329 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 353.892 ms | 0 - 1756 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 95.916 ms | 7 - 24 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® SA8775P | 122.39 ms | 0 - 1756 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® SA8775P | 122.39 ms | 0 - 1756 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® SA8775P | 122.39 ms | 0 - 1756 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® QCS9075 | 119.033 ms | 7 - 71 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 151.505 ms | 7 - 2739 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® SA7255P | 353.892 ms | 0 - 1756 MB | NPU
| proposal_generator | QNN_DLC | float | Qualcomm® SA8295P | 125.974 ms | 0 - 1430 MB | NPU
| proposal_generator | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 53.244 ms | 0 - 1510 MB | NPU
| roi_head | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 99.642 ms | 51 - 749 MB | NPU
| roi_head | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 125.168 ms | 28 - 719 MB | NPU
| roi_head | QNN_DLC | float | Snapdragon® X2 Elite | 96.983 ms | 52 - 52 MB | NPU
| roi_head | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 178.224 ms | 46 - 893 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® QCS8275 (Proxy) | 590.646 ms | 39 - 734 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® QCS8550 (Proxy) | 249.643 ms | 52 - 58 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® SA8775P | 267.888 ms | 49 - 923 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® SA8775P | 267.888 ms | 49 - 923 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® SA8775P | 267.888 ms | 49 - 923 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® QCS9075 | 342.162 ms | 52 - 106 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® QCS8450 (Proxy) | 313.188 ms | 39 - 957 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® SA7255P | 590.646 ms | 39 - 734 MB | NPU
| roi_head | QNN_DLC | float | Qualcomm® SA8295P | 307.476 ms | 49 - 846 MB | NPU
| roi_head | QNN_DLC | float | Snapdragon® 8 Elite For Galaxy Mobile | 125.168 ms | 28 - 719 MB | NPU

## License
* The license for the original implementation of MaskRCNN can be found
  [here](https://github.com/pytorch/vision/blob/main/LICENSE).

## References
* [Mask R-CNN](https://arxiv.org/abs/1703.06870)
* [Source Model Implementation](https://github.com/pytorch/vision)

## Community
* Join [our AI Hub Slack community](https://aihub.qualcomm.com/community/slack) to collaborate, post questions and learn more about on-device AI.
* For questions or feedback please [reach out to us](mailto:ai-hub-support@qti.qualcomm.com).