--- library_name: pytorch license: other tags: - foundation - android pipeline_tag: image-segmentation --- ![](https://qaihub-public-assets.s3.us-west-2.amazonaws.com/qai-hub-models/models/sam2/web-assets/model_demo.png) # 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](https://github.com/facebookresearch/sam2). 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](https://aihub.qualcomm.com/models/sam2). ### 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 | 454.419 ms | 16 - 1526 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 451.017 ms | 3 - 1456 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 368.054 ms | 17 - 1802 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 390.263 ms | 12 - 3506 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 207.751 ms | 16 - 21 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 208.088 ms | 12 - 15 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 217.912 ms | 16 - 1523 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 215.829 ms | 0 - 1454 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 454.419 ms | 16 - 1526 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 451.017 ms | 3 - 1456 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 217.912 ms | 16 - 1523 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 215.829 ms | 0 - 1454 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 154.157 ms | 2 - 1858 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 152.413 ms | 12 - 1815 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 128.163 ms | 15 - 1480 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 130.779 ms | 12 - 1446 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 95.875 ms | 16 - 1461 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Encoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 112.004 ms | 12 - 1432 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 203.577 ms | 12 - 12 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 12.372 ms | 0 - 176 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 12.357 ms | 16 - 190 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 9.597 ms | 0 - 213 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 9.492 ms | 16 - 226 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 5.273 ms | 0 - 18 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 5.282 ms | 16 - 18 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 6.659 ms | 0 - 181 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 25.03 ms | 16 - 193 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 12.372 ms | 0 - 176 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 12.357 ms | 16 - 190 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 6.659 ms | 0 - 181 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 25.03 ms | 16 - 193 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 3.692 ms | 0 - 210 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 3.696 ms | 16 - 223 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 2.923 ms | 0 - 185 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 2.923 ms | 4 - 185 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 2.578 ms | 0 - 169 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.tflite) | | SAM2Decoder | float | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 2.574 ms | 0 - 168 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Decoder | float | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 5.481 ms | 16 - 16 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2.dlc) | | SAM2Encoder | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | TFLITE | 2513.272 ms | 63 - 309 MB | CPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | QNN_DLC | 878.331 ms | 3 - 1800 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | TFLITE | 1796.779 ms | 94 - 344 MB | CPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 150.075 ms | 4 - 2098 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 142.911 ms | 3 - 1993 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 146.686 ms | 4 - 1827 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 139.531 ms | 3 - 1683 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 87.34 ms | 4 - 9 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 83.478 ms | 3 - 6 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 87.437 ms | 4 - 1548 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 83.918 ms | 3 - 1477 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 150.075 ms | 4 - 2098 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 142.911 ms | 3 - 1993 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 103.055 ms | 4 - 1541 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 95.578 ms | 3 - 1477 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 87.437 ms | 4 - 1548 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 83.918 ms | 3 - 1477 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 60.367 ms | 3 - 3634 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 56.175 ms | 3 - 1698 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 44.68 ms | 3 - 1486 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 42.662 ms | 3 - 1410 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | TFLITE | 1238.088 ms | 78 - 320 MB | CPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | QNN_DLC | 105.342 ms | 3 - 2089 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 39.721 ms | 4 - 1631 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Encoder | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 35.754 ms | 3 - 1504 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Encoder | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 83.826 ms | 3 - 3 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | TFLITE | 46.841 ms | 12 - 191 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | Dragonwing Q-6690 MTP | Qualcomm® QCM6690 | QNN_DLC | 9.911 ms | 4 - 250 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | Dragonwing RB3 Gen 2 Vision Kit | Qualcomm® QCS6490 | TFLITE | 38.801 ms | 14 - 60 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | TFLITE | 22.369 ms | 0 - 191 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | QCS8275 (Proxy) | Qualcomm® QCS8275 (Proxy) | QNN_DLC | 3.848 ms | 4 - 171 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | TFLITE | 17.787 ms | 0 - 212 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | QCS8450 (Proxy) | Qualcomm® QCS8450 (Proxy) | QNN_DLC | 2.669 ms | 4 - 198 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | TFLITE | 13.392 ms | 0 - 23 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | QCS8550 (Proxy) | Qualcomm® QCS8550 (Proxy) | QNN_DLC | 1.939 ms | 4 - 7 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | TFLITE | 14.152 ms | 0 - 190 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | QCS9075 (Proxy) | Qualcomm® QCS9075 (Proxy) | QNN_DLC | 9.341 ms | 4 - 171 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | SA7255P ADP | Qualcomm® SA7255P | TFLITE | 22.369 ms | 0 - 191 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | SA7255P ADP | Qualcomm® SA7255P | QNN_DLC | 3.848 ms | 4 - 171 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | SA8295P ADP | Qualcomm® SA8295P | TFLITE | 15.395 ms | 0 - 195 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | SA8295P ADP | Qualcomm® SA8295P | QNN_DLC | 2.897 ms | 4 - 179 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | SA8775P ADP | Qualcomm® SA8775P | TFLITE | 14.152 ms | 0 - 190 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | SA8775P ADP | Qualcomm® SA8775P | QNN_DLC | 9.341 ms | 4 - 171 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | TFLITE | 9.919 ms | 0 - 224 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | Samsung Galaxy S24 | Snapdragon® 8 Gen 3 Mobile | QNN_DLC | 1.265 ms | 4 - 195 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | TFLITE | 7.018 ms | 0 - 193 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | Samsung Galaxy S25 | Snapdragon® 8 Elite For Galaxy Mobile | QNN_DLC | 1.021 ms | 1 - 169 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | TFLITE | 16.168 ms | 14 - 186 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | Snapdragon 7 Gen 4 QRD | Snapdragon® 7 Gen 4 Mobile | QNN_DLC | 2.729 ms | 4 - 180 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | TFLITE | 5.219 ms | 0 - 207 MB | NPU | [Segment-Anything-Model-2.tflite](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.tflite) | | SAM2Decoder | w8a8 | Snapdragon 8 Elite Gen 5 QRD | Snapdragon® 8 Elite Gen 5 Mobile | QNN_DLC | 0.88 ms | 1 - 184 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | | SAM2Decoder | w8a8 | Snapdragon X Elite CRD | Snapdragon® X Elite | QNN_DLC | 2.245 ms | 4 - 4 MB | NPU | [Segment-Anything-Model-2.dlc](https://huggingface.co/qualcomm/Segment-Anything-Model-2/blob/main/Segment-Anything-Model-2_w8a8.dlc) | ## Installation Install the package via pip: ```bash # NOTE: 3.10 <= PYTHON_VERSION < 3.14 is supported. pip install "qai-hub-models[sam2]" git+https://github.com/facebookresearch/sam2.git@2b90b9f ``` ## Configure Qualcomm® AI Hub Workbench to run this model on a cloud-hosted device Sign-in to [Qualcomm® AI Hub Workbench](https://workbench.aihub.qualcomm.com/) 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. ```bash qai-hub configure --api_token API_TOKEN ``` Navigate to [docs](https://workbench.aihub.qualcomm.com/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. ```bash 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. ```bash python -m qai_hub_models.models.sam2.export ``` ## How does this work? This [export script](https://aihub.qualcomm.com/models/sam2/qai_hub_models/models/Segment-Anything-Model-2/export.py) leverages [Qualcomm® AI Hub](https://aihub.qualcomm.com/) 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. ```python 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 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. ```python 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. ```python 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](https://myaccount.qualcomm.com/signup). ## Run demo on a cloud-hosted device You can also run the demo on-device. ```bash 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](https://www.tensorflow.org/lite/android/quickstart) provides a guide to deploy the .tflite model in an Android application. - QNN (`.so` export ): This [sample app](https://docs.qualcomm.com/bundle/publicresource/topics/80-63442-50/sample_app.html) 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](https://aihub.qualcomm.com/models/sam2). Explore all available models on [Qualcomm® AI Hub](https://aihub.qualcomm.com/) ## License * The license for the original implementation of Segment-Anything-Model-2 can be found [here](https://github.com/facebookresearch/sam2/blob/main/LICENSE). ## References * [SAM 2 Segment Anything in Images and Videos](https://arxiv.org/abs/2408.00714) * [Source Model Implementation](https://github.com/facebookresearch/sam2) ## 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).