--- library_name: pytorch license: other tags: - real_time - android pipeline_tag: image-segmentation --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/yolo26_seg/web-assets/model_demo.png) # YOLO26-Segmentation: Optimized for Qualcomm Devices Ultralytics YOLO26 is a machine learning model that predicts bounding boxes, segmentation masks and classes of objects in an image. This is based on the implementation of YOLO26-Segmentation found [here](https://github.com/ultralytics/ultralytics/tree/main/ultralytics/models/yolo/segment). 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/v0.58.0/src/qai_hub_models/models/yolo26_seg) 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 Due to licensing restrictions, we cannot distribute pre-exported model assets for this model. Use the [Qualcomm® AI Hub Models](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/yolo26_seg) 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 See our repository for [YOLO26-Segmentation on GitHub](https://github.com/qualcomm/ai-hub-models/blob/v0.58.0/src/qai_hub_models/models/yolo26_seg) for usage instructions. ## Model Details **Model Type:** Model_use_case.semantic_segmentation **Model Stats:** - Model checkpoint: YOLO26N-Seg - Input resolution: 640x640 - Number of output classes: 80 - Number of parameters: 2.7M - Model size (float): 9.29 MB - Model size (w8a16): 5.25 MB ## Performance Summary | Model | Runtime | Precision | Chipset | Inference Time (ms) | Peak Memory Range (MB) | Primary Compute Unit |---|---|---|---|---|---|--- | YOLO26-Segmentation | ONNX | float | Snapdragon® X2 Elite | 4.96 ms | 17 - 17 MB | NPU | YOLO26-Segmentation | ONNX | float | Snapdragon® X Elite | 7.998 ms | 17 - 17 MB | NPU | YOLO26-Segmentation | ONNX | float | Snapdragon® 8 Gen 3 Mobile | 5.447 ms | 17 - 221 MB | NPU | YOLO26-Segmentation | ONNX | float | Snapdragon® 8 Gen 1 Mobile | 16.417 ms | 17 - 241 MB | NPU | YOLO26-Segmentation | ONNX | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 7.796 ms | 11 - 22 MB | NPU | YOLO26-Segmentation | ONNX | float | Qualcomm® QCS8450 | 16.417 ms | 17 - 241 MB | NPU | YOLO26-Segmentation | ONNX | float | Qualcomm® Dragonwing™ IQ-9075 | 9.234 ms | 15 - 18 MB | NPU | YOLO26-Segmentation | ONNX | float | Snapdragon® 8 Elite Gen 5 Mobile | 5.487 ms | 13 - 193 MB | NPU | YOLO26-Segmentation | ONNX | float | Snapdragon® 8 Elite Mobile | 5.492 ms | 12 - 191 MB | NPU | YOLO26-Segmentation | ONNX | float | Qualcomm® Dragonwing™ Q-8750 | 5.492 ms | 12 - 191 MB | NPU | YOLO26-Segmentation | ONNX | float | Qualcomm® Dragonwing™ IQ-X7181 | 7.998 ms | 17 - 17 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Snapdragon® X2 Elite | 2.69 ms | 9 - 9 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Snapdragon® X Elite | 6.458 ms | 7 - 7 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Snapdragon® 8 Gen 3 Mobile | 3.832 ms | 0 - 236 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Snapdragon® 8 Gen 1 Mobile | 8.286 ms | 8 - 246 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS6490 | 19.522 ms | 8 - 10 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 6.088 ms | 6 - 128 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Qualcomm® QCS8450 | 8.286 ms | 8 - 246 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 6.674 ms | 8 - 10 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Snapdragon® 7 Gen 4 Mobile | 7.497 ms | 8 - 223 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.493 ms | 0 - 97 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 32.377 ms | 4 - 216 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Snapdragon® 8 Elite Mobile | 2.802 ms | 0 - 96 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 7.497 ms | 8 - 223 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 2.802 ms | 0 - 96 MB | NPU | YOLO26-Segmentation | ONNX | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 6.458 ms | 7 - 7 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Snapdragon® X2 Elite | 3.172 ms | 5 - 5 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Snapdragon® X Elite | 5.811 ms | 5 - 5 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Snapdragon® 8 Gen 3 Mobile | 3.788 ms | 5 - 190 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Snapdragon® 8 Gen 1 Mobile | 10.976 ms | 5 - 207 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® QCS8275 | 16.666 ms | 0 - 158 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5.214 ms | 5 - 6 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® SA8775P | 6.854 ms | 0 - 160 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® SA8650P | 6.854 ms | 0 - 160 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® SA8255P | 6.854 ms | 0 - 160 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® QCS8450 | 10.976 ms | 5 - 207 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-9075 | 6.593 ms | 7 - 17 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.483 ms | 5 - 170 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® SA7255P | 16.666 ms | 0 - 158 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Snapdragon® 8 Elite Mobile | 3.172 ms | 5 - 167 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® SA8295P | 10.981 ms | 3 - 177 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® Dragonwing™ Q-8750 | 3.172 ms | 5 - 167 MB | NPU | YOLO26-Segmentation | QNN_DLC | float | Qualcomm® Dragonwing™ IQ-X7181 | 5.811 ms | 5 - 5 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Snapdragon® X2 Elite | 2.759 ms | 2 - 2 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Snapdragon® X Elite | 5.659 ms | 2 - 2 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Snapdragon® 8 Gen 3 Mobile | 3.537 ms | 2 - 214 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® QCS8275 | 9.508 ms | 2 - 187 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5.146 ms | 2 - 114 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® SA8775P | 5.95 ms | 0 - 187 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® SA8650P | 5.95 ms | 0 - 187 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® SA8255P | 5.95 ms | 0 - 187 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-9075 | 5.666 ms | 1 - 7 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Snapdragon® 7 Gen 4 Mobile | 5.554 ms | 2 - 190 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Snapdragon® 8 Elite Gen 5 Mobile | 2.291 ms | 2 - 196 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-6690 | 26.438 ms | 2 - 188 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® SA7255P | 9.508 ms | 2 - 187 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Snapdragon® 8 Elite Mobile | 2.571 ms | 2 - 192 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-7790 | 5.554 ms | 2 - 190 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ Q-8750 | 2.571 ms | 2 - 192 MB | NPU | YOLO26-Segmentation | QNN_DLC | w8a16 | Qualcomm® Dragonwing™ IQ-X7181 | 5.659 ms | 2 - 2 MB | NPU | YOLO26-Segmentation | TFLITE | float | Snapdragon® 8 Gen 3 Mobile | 4.105 ms | 3 - 192 MB | NPU | YOLO26-Segmentation | TFLITE | float | Snapdragon® 8 Gen 1 Mobile | 11.342 ms | 4 - 211 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® QCS8275 | 17.366 ms | 4 - 166 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® Dragonwing™ QCS8550 (Proxy) | 5.465 ms | 3 - 7 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® SA8775P | 7.047 ms | 4 - 170 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® SA8650P | 7.047 ms | 4 - 170 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® SA8255P | 7.047 ms | 4 - 170 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® QCS8450 | 11.342 ms | 4 - 211 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® Dragonwing™ IQ-9075 | 6.717 ms | 3 - 22 MB | NPU | YOLO26-Segmentation | TFLITE | float | Snapdragon® 8 Elite Gen 5 Mobile | 2.703 ms | 0 - 168 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® SA7255P | 17.366 ms | 4 - 166 MB | NPU | YOLO26-Segmentation | TFLITE | float | Snapdragon® 8 Elite Mobile | 3.194 ms | 0 - 169 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® SA8295P | 11.375 ms | 4 - 181 MB | NPU | YOLO26-Segmentation | TFLITE | float | Qualcomm® Dragonwing™ Q-8750 | 3.194 ms | 0 - 169 MB | NPU ## License * The license for the original implementation of YOLO26-Segmentation can be found [here](https://github.com/ultralytics/ultralytics/blob/main/LICENSE). ## References * [Ultralytics YOLO26 Docs: Instance Segmentation](https://docs.ultralytics.com/tasks/segment/) * [Source Model Implementation](https://github.com/ultralytics/ultralytics/tree/main/ultralytics/models/yolo/segment) ## 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).