# Image classification and encode with LiteRT The use cases use the InceptionV3 LiteRT model to classify scenes from a single camera stream and either overlay or compose the classification labels, and then encode the stream. ## Use qtivoverlay plugin to apply classification overlay Run the use case on the target device: gst-launch-1.0 -e qtiqmmfsrc name=camsrc ! video/x-raw,format=NV12,width=1280,height=720,framerate=30/1 ! queue ! tee name=split split. ! \ queue ! qtimetamux name=metamux ! queue ! qtivoverlay ! queue ! v4l2h264enc capture-io-mode=4 output-io-mode=5 ! \ h264parse ! queue ! mp4mux ! queue ! filesink location=/etc/media/output_video.mp4 split. ! queue ! qtimlvconverter ! queue ! \ qtimltflite delegate=external external-delegate-path=libQnnTFLiteDelegate.so external-delegate-options="QNNExternalDelegate,backend_type=htp;" \ model=/etc/models/inception_v3_quantized.tflite ! queue ! qtimlpostprocess settings="{\"confidence\": 40.0}" results=2 module=mobilenet-softmax \ labels=/etc/labels/classification.json ! text/x-raw ! queue ! metamux. Copy to clipboard To stop the use case, use **CTRL + C**. The following figure shows the flow of the use case execution: 1. Identify the stream coming through a camera source. 2. Overlay the classification labels using overlaylib. 3. Encode the stream as a H.264 bitstream. 4. Multiplex the stream in an MP4 container and stored as an MP4 file. **Figure : Pipeline for classification overlay and encode** The following table provides the sequential processing stages of the pipeline execution: | Process | Description | | --- | --- | | [qtiqmmfsrc](https://docs.qualcomm.com/doc/80-80021-50/topic/qtiqmmfsrc.html) |
The video stream is collected from a camera source plugin and two copies are created:
One stream is sent to the qtimetamux plugin to retain the video stream.
The other stream is sent to a ML inferencing pipeline.
Receives the video stream on its sink pad.
Color conversion
Scaling down/up
Normalization on the stream data when the model expects the floating point values as input
Converts the video stream to a tensor stream on its source pad.
The classification model uses this tensor stream for inferencing.
Loads the classification model.
Modifies the graph for the chosen delegate.
Receives the tensor stream on its sinkpad.
Runs the inference and produces a tensor stream with the inference results on its source pad.
Receives the inference tensors from a classification model on its sinkpad.
Converts the tensors into formats such as video or text that the multimedia plugins can process later.
Applies the threshold to the chosen number of results.
Loads the corresponding modules of the classification models.
In this use case, qtimlpostprocess does the following:
Loads the MobileNet-softmax submodule.
Produces results as structures of text.
Sends them to the sinkpad of qtimetamux.
Receives the video stream and text stream with classification results corresponding to video stream on its sinkpads.
Produces GST buffers with the contents of video stream on its sink pad.
Adds classification result from data sinkpad to GST buffer meta (meta muxing) on its source pad.
Receives the multiplexed stream.
Overlays the classification labels on the VideoFrame using CL.
Produces GST buffers with overlays in its source pad.
Applies parameters to each frame of the video stream it's receiving on its sinkpad.
Encodes it into bitstream and sends it over its sourcepad.
Collects the video stream (source) from a camera and creates two copies of the source:
One stream is sent to the qtimetamux plugin to retain the video stream.
The other stream is sent to an ML inferencing pipeline.
Receives the video stream on its sink pad.
Color conversion
Scaling down/up
Normalization on the stream data when the model expects the floating point values as an input
Converts the video stream to a tensor stream on its source pad.
The classification model uses the tensor stream for inferencing.
Loads the classification model.
Modifies the graph for the chosen delegate.
Receives the tensor stream on its sinkpad.
Runs the inference and produces the tensor stream with the inference results on its source pad.
Receives the inference tensors from a classification model on its sinkpad.
Converts the tensors into formats such as video or text that the multimedia plugins can process later.
Applies the threshold to the chosen number of results.
Loads the corresponding modules of the classification models.
In this use case, qtimlpostprocess does the following:
Loads Mobilenet-softmax submodule.
Produces results as video frames with classification labels.
Sends them to the sinkpad of qtivcomposer.
Receives the original video stream with classification results on its sinkpads.
On its sourcepad, produces GST buffers with contents composed of video streams from its sinkpads.
Applies parameters to each frame of the video stream it's receiving on its sinkpad.
Encodes it into bitstream and sends it over its sourcepad.