# Integrate an AI Hub model in an application This section describes how to use an AI Hub model in a reference application. This example uses an image classification model from AI Hub. Note Preprocessing and postprocessing for most of the publicly available image classification models is similar and is likely supported by the `qtimlpostprocess` plugin. Qualcomm AI Hub provides optimized AI models for Qualcomm devices, allowing them to operate on CPU, GPU, or NPU using either LiteRT (previously Tensorflow Lite) or Qualcomm® AI Engine Direct. This example explains how to use a [inception_v3](https://aihub.qualcomm.com/iot/models/inception_v3) model from AI Hub and integrate the model into the `gst-ai-classification` reference application. | **S.No** | **Reference Application** | **Verified AI Hub model** | **Label file** | | --- | --- | --- | --- | | 1 | gst-ai-classification | GoogLeNet | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 2 | gst-ai-classification | inception\_v3 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 3 | gst-ai-classification | mobilenet-v2 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 4 | gst-ai-classification | MobileNet-v3-Large | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 5 | gst-ai-classification | ResNet101 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 6 | gst-ai-classification | SqueezeNet-1\_1 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 7 | gst-ai-classification | ResNet18 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 8 | gst-ai-classification | ResNeXt50 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 9 | gst-ai-classification | WideResNet50 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 10 | gst-ai-classification | Shufflenet-v2 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 11 | gst-ai-classification | ResNeXt101 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/classification.json) | | 12 | gst-ai-segmentation | deeplabv3\_plus\_mobilenet | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/deeplabv3_resnet50.json) | | 13 | gst-ai-segmentation | fcn\_resnet50 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/deeplabv3_resnet50.json) | | 14 | gst-ai-segmentation | ffnet\_40s | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/deeplabv3_resnet50.json) | | 15 | gst-ai-segmentation | ffnet\_54s | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/deeplabv3_resnet50.json) | | 16 | gst-ai-segmentation | ffnet\_78s | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/deeplabv3_resnet50.json) | | 17 | gst-ai-object-detection | Yolo-v7 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/yolox.json) | | 18 | gst-ai-object-detection | YOLOv8-Detection | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/yolox.json) | | 19 | gst-ai-superresolution | QuickSRNetLarge | Not applicable | | 20 | gst-ai-superresolution | QuickSRNetMedium | Not applicable | | 21 | gst-ai-superresolution | QuickSRNetSmall | Not applicable | | 22 | gst-ai-superresolution | XLSR | Not applicable | | 23 | gst-ai-monodepth | Midas-V2 | [Labels](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/monodepth.json) | | 24 | gst-ai-pose-detection | HRNetPose | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | | Note Some models only have LiteRT or Qualcomm AI Engine Direct SDK models available with AI Hub. However, AI Hub frequently updates its library to include more models and enhance existing ones. Prerequisites 1. Download [AI Hub models](https://aihub.qualcomm.com/iot/models/inception_v3). 1. Search for the model in the search bar and select the model. 2. Select the runtime, precision, and device for which you want to download the model. - Runtime: *TFLite* - Precision: *w8a8* - Device: - Qualcomm Dragonwing™ RB3 Gen 2: *Qualcomm QCS6490* - Qualcomm Dragonwing™ IQ-9075: *Qualcomm QCS9075* 2. Copy the label file that matches the model to the device. > > > Label files are available for download at the provided > [links](https://github.com/quic/sample-apps-for-qualcomm-linux/blob/main/artifacts/json_labels/). Note The following sections use placeholder model names for models downloaded through AI Hub. Update the model name in the commands with the appropriate model name as downloaded through AI Hub. ## Modify the configuration 1. Copy the model to the device. scp inception_v3-inception-v3-w8a8.tflite root@:/etc/models/ Copy to clipboard 2. Copy the labels to the device. wget https://raw.githubusercontent.com/quic/sample-apps-for-qualcomm-linux/refs/heads/main/artifacts/json_labels/classification.json Copy to clipboard scp classification.json root@:/etc/labels/ Copy to clipboard 3. Update the `/etc/configs/config_classification.json` file as shown below. { "file-path": "/etc/media/video.mp4", "ml-framework": "tflite", "model": "/etc/models/inception_v3-inception-v3-w8a8.tflite", "labels": "/etc/labels/classification.json", "threshold": 40, "runtime": "dsp" } Copy to clipboard ## Run the sample application 1. Enable the Weston service for the display by running the following command: export XDG_RUNTIME_DIR=/dev/socket/weston && export WAYLAND_DISPLAY=wayland-1 Copy to clipboard 2. Run the reference application by giving the constants values in the command line parameters: gst-ai-classification --config-file=/etc/configs/config_classification.json Copy to clipboard 3. Use sample applications to run various AI Hub models on your devices. To run different AI Hub models: 1. Download the model and label files that you want to run. 2. Copy the model and label files to the device. 3. Run the sample application by following the steps below. Tab gst-ai-object-detection Tab gst-ai-segmentation Tab gst-ai-superresolution 1. Download the [Yolox Quantized LiteRT model file](https://aihub.qualcomm.com/iot/models/yolox) (used for object detection). 2. Update the `/etc/configs/config_detection.json` file as shown below. { "file-path": "/etc/media/video.mp4", "ml-framework": "tflite", "yolo-model-type": "yolox", "model": "/etc/models/yolox_quantized.tflite", "labels": "/etc/labels/yolox.json", "threshold": 40, "runtime": "dsp" } Copy to clipboard Note Update the model name in the configuration file according to the model you obtained. 3. Run the sample app. Note Ensure that you have copied the model and label files to the device. gst-ai-object-detection --config-file=/etc/configs/config_detection.json Copy to clipboard 1. Update the `/etc/configs/config_segmentation.json` file as follows: Take the `deeplabv3_plus_mobilenet_quantized.tflite` model file (used for segmentation) and the `deeplabv3_resnet50.json` file as an example. { "file-path": "/etc/media/video.mp4", "ml-framework": "tflite", "model": "/etc/models/deeplabv3_plus_mobilenet_quantized.tflite", "labels": "/etc/labels/deeplabv3_resnet50.json", "runtime": "dsp" } Copy to clipboard 2. Run the sample app. gst-ai-segmentation --config-file=/etc/configs/config_segmentation.json Copy to clipboard 1. Update the `/etc/configs/config-superresolution.json` file as shown below. { "input-file-path": "/etc/media/super.mp4", "model": "/etc/models/quicksrnetsmall_quantized.tflite", "output-file-path":"/etc/media/out.mp4" } Copy to clipboard 2. Run the sample app: gst-ai-superresolution --config-file=/etc/configs/config-superresolution.json Copy to clipboard Note Ensure that the video used has a 128x128 resolution, which is the input to the model. For additional details regarding the input parameters, see the help section of the application. Note In cases of segmentation, objects that aren’t recognized may appear against a white background. This issue stems from the label file and will be addressed in future updates. 4. Recompile the reference application if you want to make any changes and run it. See the [Qualcomm Intelligent Multimedia Software Development Kit (IM SDK) Quick Start Guide](https://docs.qualcomm.com/bundle/publicresource/topics/80-70022-51/introduction.html) for instructions on how to compile reference applications with the eSDK. Last Published: Nov 28, 2025 [Previous Topic Integrate a custom AI model in an application](https://docs.qualcomm.com/bundle/publicresource/80-70022-15B/topics/integrate-custom-model.md) [Next Topic Add postprocessing support for a custom model](https://docs.qualcomm.com/bundle/publicresource/80-70022-15B/topics/add-postprocessing-support-custom-model.md) Source: [https://docs.qualcomm.com/doc/80-70022-15B/topic/integrate-ai-hub-models.html](https://docs.qualcomm.com/doc/80-70022-15B/topic/integrate-ai-hub-models.html)