RK人工智能算法最全官方示例
RKNN Model Zoo基于 RKNPU SDK 工具链开发, 提供了目前主流算法的部署例程. 例程包含导出RKNN模型, 使用 Python API, CAPI 推理 RKNN 模型的流程.[*]支持 RK3562, RK3566, RK3568, RK3576, RK3588, RV1126B 平台。
[*]部分支持RV1103, RV1106
[*]支持 RV1109, RV1126, RK1808 平台。
源文地址：https://github.com/airockchip/rknn_model_zoo/tree/main
以下demo除了从对应的仓库导出模型, 也可从网盘 https://console.zbox.filez.com/l/8ufwtG (提取码: rknn) 下载模型文件.
CategoryNameDtypeModel Download LinkSupport platform
图像分类mobilenetFP16/INT8mobilenetv2-12.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RV1103|RV1106
RK1808|RK3399PRO
RV1109|RV1126
图像分类resnetFP16/INT8resnet50-v2-7.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
物体检测yolov5FP16/INT8./yolov5s_relu.onnx
./yolov5n.onnx
./yolov5s.onnx
./yolov5m.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RV1103|RV1106
RK1808|RK3399PRO
RV1109|RV1126
物体检测yolov6FP16/INT8./yolov6n.onnx
./yolov6s.onnx
./yolov6m.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
物体检测yolov7FP16/INT8./yolov7-tiny.onnx
./yolov7.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
物体检测yolov8FP16/INT8./yolov8n.onnx
./yolov8s.onnx
./yolov8m.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
物体检测yolov8_obbINT8./yolov8n-obb.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
物体检测yolov10FP16/INT8./yolov10n.onnx
./yolov10s.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RV1103|RV1106
RK1808|RK3399PRO
RV1109|RV1126
物体检测yolo11FP16/INT8./yolo11n.onnx
./yolo11s.onnx
./yolo11m.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RV1103|RV1106
RK1808|RK3399PRO
RV1109|RV1126
物体检测yoloxFP16/INT8./yolox_s.onnx
./yolox_m.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
物体检测ppyoloeFP16/INT8./ppyoloe_s.onnx
./ppyoloe_m.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
物体检测yolo_worldFP16/INT8./yolo_world_v2s.onnx
./clip_text.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
人体关键点yolov8_poseINT8./yolov8n-pose.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
图像分割deeplabv3FP16/INT8./deeplab-v3-plus-mobilenet-v2.pbRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
图像分割yolov5_segFP16/INT8./yolov5n-seg.onnx
./yolov5s-seg.onnx
./yolov5m-seg.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
图像分割yolov8_segFP16/INT8./yolov8n-seg.onnx
./yolov8s-seg.onnx
./yolov8m-seg.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
图像分割ppsegFP16/INT8pp_liteseg_cityscapes.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
图像分割mobilesamFP16mobilesam_encoder_tiny.onnx
mobilesam_decoder.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
人脸关键点RetinaFaceINT8RetinaFace_mobile320.onnx
RetinaFace_resnet50_320.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
车牌识别LPRNetFP16/INT8./lprnet.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RV1103|RV1106
RK1808|RK3399PRO
RV1109|RV1126
文字检测PPOCR-DetFP16/INT8../ppocrv4_det.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
文字识别PPOCR-RecFP16../ppocrv4_rec.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
自然语言翻译lite_transformerFP16lite-transformer-encoder-16.onnx
lite-transformer-decoder-16.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
RK1808|RK3399PRO
RV1109|RV1126
图文匹配clipFP16./clip_images.onnx
./clip_text.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
语音识别wav2vec2FP16wav2vec2_base_960h_20s.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
语音识别whisperFP16whisper_encoder_base_20s.onnx
whisper_decoder_base_20s.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
语音识别zipformerFP16encoder-epoch-99-avg-1.onnx
decoder-epoch-99-avg-1.onnx
joiner-epoch-99-avg-1.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
语音分类yamnetFP16yamnet_3s.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
文字转语音mms_ttsFP16mms_tts_eng_encoder_200.onnx
mms_tts_eng_decoder_200.onnxRK3562|RK3566|RK3568|RK3576|RK3588|RV1126B
Model performance benchmark(FPS)
demomodel_nameinputs_shape dtypeRK3566
RK3568RK3562RK3588
@single_coreRK3576
@single_coreRV1109RV1126RK1808
mobilenetmobilenetv2-12INT8180.7281.3450.7467.0212.9322.3170.3
resnetresnet50-v2-7INT837.954.9110.199.024.436.237.1
yolov5yolov5s_reluINT825.533.266.165.020.229.237.2
yolov5nINT839.747.482.5112.736.353.261.2
yolov5sINT819.323.648.457.513.620.028.2
yolov5mINT88.610.820.923.75.88.513.3
yolov6yolov6nINT848.856.4106.4109.137.856.866.8
yolov6sINT815.217.336.435.010.816.324.1
yolov6mINT87.28.617.817.45.68.311.5
yolov7yolov7-tinyINT827.936.572.774.815.422.437.2
yolov7INT84.65.911.413.03.34.87.4
yolov8yolov8nINT834.040.973.590.224.035.442.3
yolov8sINT815.118.438.040.88.913.119.1
yolov8mINT86.58.216.216.73.95.89.1
yolov8_obbyolov8n-obbINT833.941.374.090.225.137.342.8
yolov10yolov10nINT820.734.161.280.2///
yolov10sINT810.316.933.839.9///
yolo11yolo11nINT820.634.060.077.911.717.017.6
yolo11sINT810.216.733.038.25.07.38.4
yolo11mINT84.66.512.714.62.84.05.1
yoloxyolox_sINT815.218.337.141.510.615.723.0
yolox_mINT86.68.216.017.64.66.810.7
ppyoloeppyoloe_sINT817.120.032.541.311.216.421.1
ppyoloe_mINT87.89.215.817.85.27.79.4
yolo_worldyolo_world_v2sINT87.49.622.122.3///
clip_textFP1629.867.495.863.5///
yolov8_poseyolov8n-poseINT822.631.055.966.8///
deeplabv3deeplab-v3-plus-mobilenet-v2INT810.921.434.039.410.113.04.4
yolov5_segyolov5n-segINT832.238.569.388.328.642.249.6
yolov5s-segINT815.018.136.841.69.614.022.5
yolov5m-segINT86.88.416.418.04.76.810.8
yolov8_segyolov8n-segINT827.833.060.871.118.627.632.9
yolov8s-segINT811.714.128.930.86.69.814.6
yolov8m-segINT85.26.412.612.73.14.66.9
ppsegppseg_lite_1024x512INT85.913.935.733.618.427.120.9
mobilesammobilesam_encoder_tinyFP161.06.610.011.9///
mobilesam_decoderFP1624.369.6116.4108.6///
RetinaFaceRetinaFace_mobile320INT8156.4300.8227.2470.5144.8212.5198.5
RetinaFace_resnet50_320INT818.726.949.256.614.620.824.6
LPRNetlprnetFP16143.2420.6586.4647.830.6(INT8)47.6(INT8)30.1(INT8)
PPOCR-Detppocrv4_detINT822.128.050.764.311.016.114.2
PPOCR-Recppocrv4_recFP1619.554.373.996.81.01.66.7
lite_transformerlite-transformer-encoder-16embedding-256, token-16FP16337.5725.8867.6784.122.735.498.3
lite-transformer-decoder-16embedding-256, token-16FP16142.5252.0343.8272.348.065.8109.9
clipclip_imagesFP162.33.46.56.7///
clip_textFP1629.766.696.063.7///
wav2vec2wav2vec2_base_960h_20s20s audioFP16RTF
0.817RTF
0.323RTF
0.133RTF
0.073///
whisperwhisper_base_20s20s audioFP16RTF
1.178RTF
0.420RTF
0.215RTF
0.218///
zipformerzipformer-bilingual-zh-en-tstreaming audioFP16RTF
0.196RTF
0.116RTF
0.065RTF
0.082///
yamnetyamnet_3s3s audioFP16RTF
0.013RTF
0.008RTF
0.004RTF
0.005///
mms_ttsmms_tts_eng_200token-200FP16RTF
0.311RTF
0.138RTF
0.069RTF
0.069///
[*]该性能数据基于各平台的最大NPU频率进行测试
[*]该性能数据指模型推理的耗时, 不包含前后处理的耗时
[*]/表示当前版本暂不支持
Demo编译说明
对于 Linux 系统的开发板:./build-linux.sh -t <target> -a <arch> -d <build_demo_name> [-b <build_type>] [-m]
-t : target (rk356x/rk3576/rk3588/rv1106/rv1126b/rv1126/rk1808)
-a : arch (aarch64/armhf)
-d : demo name
-b : build_type(Debug/Release)
-m : enable address sanitizer, build_type need set to Debug
Note: 'rk356x' represents rk3562/rk3566/rk3568, 'rv1106' represents rv1103/rv1106, 'rv1126' represents rv1109/rv1126，'rv1126b' is different from 'rv1126'.
# 以编译64位Linux RK3566的yolov5 demo为例:
./build-linux.sh -t rk356x -a aarch64 -d yolov5
对于 Android 系统的开发板: # 对于 Android 系统的开发板, 首先需要根据实际情况, 设置安卓NDK编译工具的路径
<font color="#ff0000">export</font> ANDROID_NDK_PATH=~/opts/ndk/android-ndk-r18b
./build-android.sh -t <target> -a <arch> -d <build_demo_name> [-b <build_type>] [-m]
-t : target (rk356x/rk3588/rk3576)
-a : arch (arm64-v8a/armeabi-v7a)
-d : demo name
-b : build_type (Debug/Release)
-m : enable address sanitizer, build_type need set to Debug
# 以编译64位Android RK3566的yolov5 demo为例:
./build-android.sh -t rk356x -a arm64-v8a -d yolov5
版本RKNPU2 SDKRKNPU1 SDK
2.3.2>=2.3.2>=1.7.5
2.3.0>=2.3.0>=1.7.5
2.2.0>=2.2.0>=1.7.5
2.1.0>=2.1.0>=1.7.5
2.0.0>=2.0.0>=1.7.5
1.6.0>=1.6.0-
1.5.0>=1.5.0>=1.7.3
RKNPU相关资料
[*]RKNPU2 SDK: https://github.com/airockchip/rknn-toolkit2
[*]RKNPU1 SDK: https://github.com/airockchip/rknn-toolkit
页:
[1]