# 使用 Neural Processing SDK 进行图像分割和编码

Source: [https://docs.qualcomm.com/doc/80-70017-50SC/topic/single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized.html](https://docs.qualcomm.com/doc/80-70017-50SC/topic/single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized.html)

该用例使用 DeepLab v3 模型和 Qualcomm Neural Processing SDK runtime 来合成语义分割和原始视频流，对该数据流进行编码，接着在 MP4 容器中对其进行多路复用。

运行用例：

    gst-launch-1.0 -e --gst-debug=2 \
    qtiqmmfsrc name=camsrc ! video/x-raw\(memory:GBM\),format=NV12,width=1920,height=1080,framerate=30/1,compression=ubwc ! queue ! tee name=split \
    split. ! queue ! qtivcomposer name=mixer sink_1::dimensions="<1920,1080>" sink_1::alpha=0.5 ! queue ! video/x-raw\(memory:GBM\),format=NV12,width=1920,height=1080,interlace-mode=progressive,colorimetry=bt601 ! v4l2h264enc capture-io-mode=5 output-io-mode=5 ! h264parse ! queue ! mp4mux ! queue ! filesink location=/opt/video.mp4 \
    split. ! queue ! qtimlvconverter ! queue ! qtimlsnpe delegate=dsp model=/opt/deeplabv3_resnet50.dlc ! queue ! qtimlvsegmentation module=deeplab-argmax labels=/opt/deeplabv3_resnet50.labels ! video/x-raw,width=640,height=360 ! queue ! mixer.Copy to clipboard

如需停止用例，可按下 CTRL + C。

Figure : 使用 qtivcomposer 进行图像分割和编码的 pipeline
            
            ![](data:image/png;base64,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)

下图显示了用例执行流程：

1. 识别从摄像头源传来的视频流中的场景。
2. 使用 qtivcomposer 合成语义分割和视频流。
3. 将流编码为 H.264 码流并在 MP4 容器中多路复用该流。

pipeline 执行的顺序处理阶段如下表所示：

| 处理过程 | 说明 |
| --- | --- |
| [qtiqmmfsrc](https://docs.qualcomm.com/doc/80-70017-50SC/topic/qtiqmmfsrc.html) | <ol class="ol" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ol_f5k_g5n_vbc"><br>                                <li class="li">采集视频流（源）并创建源的两个副本：<ul class="ul" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ul_n44_nwl_vbc"><br>                                        <li class="li">一个视频流被发送到 qtivcomposer 插件以保留视频流。</li><br><br>                                        <li class="li">另一个流被发送到 pipeline 中的 ML 推理分支。</li><br><br>                                    </ul><br></li><br><br>                            </ol> |
| **预处理** | **预处理** |
| [qtimlvconverter](https://docs.qualcomm.com/doc/80-70017-50SC/topic/qtimlvconverter.html) | <ol class="ol" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ol_xsf_q5l_vbc"><br>                                <li class="li">在其接收端上接收视频流。</li><br><br>                                <li class="li">执行预处理：<ul class="ul" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ul_ff2_twl_vbc"><br>                                        <li class="li">颜色转换</li><br><br>                                        <li class="li">缩小/放大</li><br><br>                                        <li class="li">在模型需要浮点值作为输入时，对流数据进行归一化</li><br><br>                                    </ul><br></li><br><br>                                <li class="li">在其发送端上将视频流转换为张量数据。<p class="p">分割模型使用此张量数据进行推理。</p><br></li><br><br>                            </ol> |
| **推理** | **推理** |
| [qtimlsnpe](https://docs.qualcomm.com/doc/80-70017-50SC/topic/qtimlsnpe.html) | <ol class="ol" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ol_lfr_35n_vbc"><br>                                <li class="li">加载分割模型。</li><br><br>                                <li class="li">为选择的 delegate 修改图。</li><br><br>                                <li class="li">在其接收端上接收张量数据。</li><br><br>                                <li class="li">运行推理并在其发送端上生成包含分割结果的张量数据。</li><br><br>                            </ol> |
| **后处理** | **后处理** |
| [qtimlvsegmentation](https://docs.qualcomm.com/doc/80-70017-50SC/topic/qtimlvsegmentation.html) | <ol class="ol" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ol_mtr_k5n_vbc"><br>                                <li class="li">在其接收端上接收推理张量。</li><br><br>                                <li class="li">将推理张量转换为多媒体插件稍后可以处理的视频格式。</li><br><br>                                <li class="li">生成帧的语义分割。</li><br><br>                                <li class="li">加载分割模型的相应模块。<p class="p">在此用例中，qtimlvsegmentation 执行以下操作： </p><ol class="ol" type="a" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ol_ntr_k5n_vbc"><br>                                        <li class="li">加载 deeplab-argmax 子模块。</li><br><br>                                        <li class="li">生成带有分割掩码的视频帧。</li><br><br>                                        <li class="li">将它们发送至 qtivcomposer 的接收端。</li><br><br>                                    </ol><br><br>                                </li><br><br>                            </ol> |
| [qtivcomposer](https://docs.qualcomm.com/doc/80-70017-50SC/topic/qtivcomposer.html) | <ol class="ol" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ol_nmc_lxl_vbc"><br>                                <li class="li">在其接收端上接收带有分割掩码的原始视频流。 </li><br><br>                                <li class="li">在其发送端上生成 GST 缓存，其内容由来自其接收端的视频流组成。</li><br><br>                            </ol> |
| [v4l2h264enc](https://docs.qualcomm.com/doc/80-70017-50SC/topic/v4l2h264enc.html) | <ol class="ol" id="single-camera-stream-with-image-segmentation-and-encode-with-deeplabv3-quantized__ol_wsc_bsn_vbc"><br>                                <li class="li">将参数应用于在接收端上接收的视频流的每一帧。</li><br><br>                                <li class="li">将其编码为码流，并通过其发送端发送。</li><br><br>                            </ol> |
| h264parse | 将与码流对应的其他信息添加到 GStreamer 缓存元数据。 |
| mp4mux | 接收这些缓存并创建具有格式规范缓存的容器。 |
| **输出** | **输出** |
| Filesink | 将生成的数据流存储在 /opt/video.mp4 文件中。 |
| 播放 | 从主机拉出 video.mp4 并在媒体播放器上播放：<br>`scp root@<IP address of target<br>                                    device>:/opt/ <destination directory>` |

**上一级主题：** [Qualcomm Neural Processing SDK 用例](https://docs.qualcomm.com/doc/80-70017-50SC/topic/qualcomm-neural-processing-sdk-use-cases.html)

Last Published: Nov 11, 2025

[Previous Topic
使用 Neural Processing SDK 进行图像分割和显示](https://docs.qualcomm.com/bundle/publicresource/80-70017-50SC/topics/single-camera-stream-with-image-segmentation-and-display-with-deeplabv3-quantized.md) [Next Topic
多媒体用例](https://docs.qualcomm.com/bundle/publicresource/80-70017-50SC/topics/multimedia-use-cases.md)