文章目录
- 前言
- 1. 构建python环境
- 2 模型转换
前言
我是一个开发板的新手，刚买了一个瑞芯微3588s的板子，目标是要学习嵌入式的开发，也就是说把深度学习的框架，跑到板子上。万丈高楼平地起步。先把仿真环境搭建起来。
仿真环境可以跑在ubuntu的机器上，但仿真环境只能运行python代码，而真是的3588s的板子是既能跑c/c++代码又能跑python代码。
1. 构建python环境
安装python环境
conda create -n rknnpy38 python=3.8
conda activate rknnpy38
git 仓库：https://github.com/rockchip-linux/rknn-toolkit2
到这里下载代码
614 cd /home/ubuntu/Downloads/rknn-toolkit2-master/doc
615 pip install -r requirements_cp38-1.5.0.txt
cd /home/ubuntu/Downloads/rknn-toolkit2-master/examples/onnx/yolov5
rknnpy38) ubuntu@ubuntu:~/Downloads/rknn-toolkit2-master/examples/onnx/yolov5$ python test.py
W __init__: rknn-toolkit2 version: 1.5.0+1fa95b5c
--> Config model
W config: 'target_platform' is None, use rk3566 as default, Please set according to the actual platform!
done
--> Loading model
Loading : 100%|████████████████████████████████████████████████| 124/124 [00:00<00:00, 26012.49it/s]
done
--> Building model
I base_optimize ...
I base_optimize done.
I
I fold_constant ...
I fold_constant done.
I
I correct_ops ...
I correct_ops done.
I
I fuse_ops ...
I fuse_ops results:
I convert_resize_to_deconv: remove node = ['Resize_84'], add node = ['Resize_84_2deconv']
I convert_resize_to_deconv: remove node = ['Resize_100'], add node = ['Resize_100_2deconv']
I fold_constant ...
I fold_constant done.
I fuse_ops done.
I
I sparse_weight ...
I sparse_weight done.
I
Analysing : 100%|███████████████████████████████████████████████| 142/142 [00:00<00:00, 6658.07it/s]
Quantizating : 100%|█████████████████████████████████████████████| 142/142 [00:01<00:00, 113.40it/s]
I
I quant_optimizer ...
I quant_optimizer results:
I adjust_relu: ['Relu_140', 'Relu_137', 'Relu_135', 'Relu_133', 'Relu_131', 'Relu_128', 'Relu_126', 'Relu_123', 'Relu_121', 'Relu_119', 'Relu_117', 'Relu_114', 'Relu_112', 'Relu_109', 'Relu_107', 'Relu_105', 'Relu_103', 'Relu_98', 'Relu_96', 'Relu_93', 'Relu_91', 'Relu_89', 'Relu_87', 'Relu_82', 'Relu_80', 'Relu_74', 'Relu_72', 'Relu_69', 'Relu_66', 'Relu_64', 'Relu_62', 'Relu_60', 'Relu_58', 'Relu_55', 'Relu_52', 'Relu_50', 'Relu_47', 'Relu_45', 'Relu_42', 'Relu_40', 'Relu_38', 'Relu_36', 'Relu_34', 'Relu_31', 'Relu_28', 'Relu_26', 'Relu_23', 'Relu_21', 'Relu_19', 'Relu_17', 'Relu_15', 'Relu_12', 'Relu_9', 'Relu_7', 'Relu_5', 'Relu_3', 'Relu_1']
I adjust_no_change_node: ['MaxPool_77']
I quant_optimizer done.
I
W build: The default input dtype of 'images' is changed from 'float32' to 'int8' in rknn model for performance!
Please take care of this change when deploy rknn model with Runtime API!
W build: The default output dtype of '269' is changed from 'float32' to 'int8' in rknn model for performance!
Please take care of this change when deploy rknn model with Runtime API!
W build: The default output dtype of '271' is changed from 'float32' to 'int8' in rknn model for performance!
Please take care of this change when deploy rknn model with Runtime API!
W build: The default output dtype of '273' is changed from 'float32' to 'int8' in rknn model for performance!
Please take care of this change when deploy rknn model with Runtime API!
I rknn building ...
I RKNN: [23:18:44.124] compress = 0, conv_eltwise_activation_fuse = 1, global_fuse = 1, multi-core-model-mode = 7, output_optimize = 1,enable_argb_group=0
I RKNN: librknnc version: 1.5.0 (e6fe0c678@2023-05-25T16:15:03)
D RKNN: [23:18:44.163] RKNN is invoked
D RKNN: [23:18:44.244] >>>>>> start: N4rknn19RKNNSetOpTargetPassE
D RKNN: [23:18:44.244] <<<<<<<< end: N4rknn19RKNNSetOpTargetPassE
D RKNN: [23:18:44.244] >>>>>> start: N4rknn16RKNNAddFirstConvE
D RKNN: [23:18:44.244] <<<<<<<< end: N4rknn16RKNNAddFirstConvE
D RKNN: [23:18:44.244] >>>>>> start: N4rknn27RKNNEliminateQATDataConvertE
D RKNN: [23:18:44.244] <<<<<<<< end: N4rknn27RKNNEliminateQATDataConvertE
D RKNN: [23:18:44.244] >>>>>> start: N4rknn17RKNNTileGroupConvE
D RKNN: [23:18:44.244] <<<<<<<< end: N4rknn17RKNNTileGroupConvE
D RKNN: [23:18:44.244] >>>>>> start: N4rknn19RKNNTileFcBatchFuseE
D RKNN: [23:18:44.244] <<<<<<<< end: N4rknn19RKNNTileFcBatchFuseE
D RKNN: [23:18:44.244] >>>>>> start: N4rknn15RKNNAddConvBiasE
D RKNN: [23:18:44.245] <<<<<<<< end: N4rknn15RKNNAddConvBiasE
D RKNN: [23:18:44.245] >>>>>> start: N4rknn15RKNNTileChannelE
D RKNN: [23:18:44.245] <<<<<<<< end: N4rknn15RKNNTileChannelE
D RKNN: [23:18:44.245] >>>>>> start: N4rknn18RKNNPerChannelPrepE
D RKNN: [23:18:44.245] <<<<<<<< end: N4rknn18RKNNPerChannelPrepE
D RKNN: [23:18:44.245] >>>>>> start: N4rknn11RKNNBnQuantE
D RKNN: [23:18:44.245] <<<<<<<< end: N4rknn11RKNNBnQuantE
D RKNN: [23:18:44.245] >>>>>> start: N4rknn21RKNNFuseOptimizerPassE
D RKNN: [23:18:44.246] <<<<<<<< end: N4rknn21RKNNFuseOptimizerPassE
D RKNN: [23:18:44.246] >>>>>> start: N4rknn15RKNNTurnAutoPadE
D RKNN: [23:18:44.246] <<<<<<<< end: N4rknn15RKNNTurnAutoPadE
D RKNN: [23:18:44.246] >>>>>> start: N4rknn16RKNNInitRNNConstE
D RKNN: [23:18:44.246] <<<<<<<< end: N4rknn16RKNNInitRNNConstE
D RKNN: [23:18:44.246] >>>>>> start: N4rknn17RKNNInitCastConstE
D RKNN: [23:18:44.246] <<<<<<<< end: N4rknn17RKNNInitCastConstE
D RKNN: [23:18:44.246] >>>>>> start: N4rknn20RKNNMultiSurfacePassE
D RKNN: [23:18:44.246] <<<<<<<< end: N4rknn20RKNNMultiSurfacePassE
D RKNN: [23:18:44.246] >>>>>> start: OpEmit
D RKNN: [23:18:44.247] <<<<<<<< end: OpEmit
D RKNN: [23:18:44.247] >>>>>> start: N4rknn19RKNNLayoutMatchPassE
D RKNN: [23:18:44.247] >>>>>> start: N4rknn20RKNNAddSecondaryNodeE
D RKNN: [23:18:44.247] <<<<<<<< end: N4rknn20RKNNAddSecondaryNodeE
D RKNN: [23:18:44.247] >>>>>> start: OpEmit
D RKNN: [23:18:44.250] <<<<<<<< end: OpEmit
D RKNN: [23:18:44.250] >>>>>> start: N4rknn23RKNNProfileAnalysisPassE
D RKNN: [23:18:44.250] <<<<<<<< end: N4rknn23RKNNProfileAnalysisPassE
D RKNN: [23:18:44.251] >>>>>> start: N4rknn21RKNNOperatorIdGenPassE
D RKNN: [23:18:44.251] <<<<<<<< end: N4rknn21RKNNOperatorIdGenPassE
D RKNN: [23:18:44.251] >>>>>> start: N4rknn23RKNNWeightTransposePassE
D RKNN: [23:18:44.385] <<<<<<<< end: N4rknn23RKNNWeightTransposePassE
D RKNN: [23:18:44.385] >>>>>> start: N4rknn26RKNNCPUWeightTransposePassE
D RKNN: [23:18:44.385] <<<<<<<< end: N4rknn26RKNNCPUWeightTransposePassE
D RKNN: [23:18:44.385] >>>>>> start: N4rknn18RKNNModelBuildPassE
D RKNN: [23:18:44.617] RKNNModelBuildPass: [Statistics]
D RKNN: [23:18:44.617] total_regcfg_size : 299856
D RKNN: [23:18:44.617] total_diff_regcfg_size: 164496
D RKNN: [23:18:44.618] ID OpType DataType Target InputShape OutputShape DDR Cycles NPU Cycles Total Cycles Time(us) MacUsage(%) Task Number Task Size Regcmd Size RW(KB) FullName
D RKNN: [23:18:44.618] 0 InputOperator INT8 CPU \ (1,3,640,640) 0 0 0 0 \ 0 0 0 1200.00 InputOperator:images
D RKNN: [23:18:44.618] 1 ConvRelu INT8 NPU (1,3,640,640),(32,3,6,6),(32) (1,32,320,320) 0 0 0 0 \ 16 9440 10240 4409.25 Conv:Conv_0
D RKNN: [23:18:44.618] 2 ConvRelu INT8 NPU (1,32,320,320),(64,32,3,3),(64) (1,64,160,160) 0 0 0 0 \ 16 4664 5760 4818.50 Conv:Conv_2
D RKNN: [23:18:44.618] 3 ConvRelu INT8 NPU (1,64,160,160),(32,64,1,1),(32) (1,32,160,160) 0 0 0 0 \ 8 3208 3712 2402.25 Conv:Conv_4
D RKNN: [23:18:44.618] 4 ConvRelu INT8 NPU (1,32,160,160),(32,32,1,1),(32) (1,32,160,160) 0 0 0 0 \ 4 2504 2688 1601.25 Conv:Conv_6
D RKNN: [23:18:44.618] 5 ConvReluAdd INT8 NPU (1,32,160,160),(32,32,3,3),(32),(1,32,160,160) (1,32,160,160) 0 0 0 0 \ 4 2560 2688 2409.25 Conv:Conv_8
D RKNN: [23:18:44.618] 6 ConvRelu INT8 NPU (1,64,160,160),(32,64,1,1),(32) (1,32,160,160) 0 0 0 0 \ 8 3208 3712 2402.25 Conv:Conv_11
D RKNN: [23:18:44.618] 7 Concat INT8 NPU (1,32,160,160),(1,32,160,160) (1,64,160,160) 0 0 0 0 \ 2 1104 1280 3200.00 Concat:Concat_13
D RKNN: [23:18:44.618] 8 ConvRelu INT8 NPU (1,64,160,160),(64,64,1,1),(64) (1,64,160,160) 0 0 0 0 \ 8 3208 3712 3204.50 Conv:Conv_14
D RKNN: [23:18:44.618] 9 ConvRelu INT8 NPU (1,64,160,160),(128,64,3,3),(128) (1,128,80,80) 0 0 0 0 \ 8 3256 3712 2473.00 Conv:Conv_16
D RKNN: [23:18:44.618] 10 ConvRelu INT8 NPU (1,128,80,80),(64,128,1,1),(64) (1,64,80,80) 0 0 0 0 \ 4 2504 2688 1208.50 Conv:Conv_18
D RKNN: [23:18:44.618] 11 ConvRelu INT8 NPU (1,64,80,80),(64,64,1,1),(64) (1,64,80,80) 0 0 0 0 \ 2 2112 2176 804.50 Conv:Conv_20
D RKNN: [23:18:44.618] 12 ConvReluAdd INT8 NPU (1,64,80,80),(64,64,3,3),(64),(1,64,80,80) (1,64,80,80) 0 0 0 0 \ 2 2112 2176 1236.50 Conv:Conv_22
D RKNN: [23:18:44.618] 13 ConvRelu INT8 NPU (1,64,80,80),(64,64,1,1),(64) (1,64,80,80) 0 0 0 0 \ 2 2112 2176 804.50 Conv:Conv_25
D RKNN: [23:18:44.618] 14 ConvReluAdd INT8 NPU (1,64,80,80),(64,64,3,3),(64),(1,64,80,80) (1,64,80,80) 0 0 0 0 \ 2 2112 2176 1236.50 Conv:Conv_27
D RKNN: [23:18:44.618] 15 ConvRelu INT8 NPU (1,128,80,80),(64,128,1,1),(64) (1,64,80,80) 0 0 0 0 \ 4 2504 2688 1208.50 Conv:Conv_30
D RKNN: [23:18:44.618] 16 Concat INT8 NPU (1,64,80,80),(1,64,80,80) (1,128,80,80) 0 0 0 0 \ 2 1104 1280 1600.00 Concat:Concat_32
D RKNN: [23:18:44.618] 17 ConvRelu INT8 NPU (1,128,80,80),(128,128,1,1),(128) (1,128,80,80) 0 0 0 0 \ 4 2504 2688 1617.00 Conv:Conv_33
D RKNN: [23:18:44.618] 18 ConvRelu INT8 NPU (1,128,80,80),(256,128,3,3),(256) (1,256,40,40) 0 0 0 0 \ 4 2552 2688 1490.00 Conv:Conv_35
D RKNN: [23:18:44.618] 19 ConvRelu INT8 NPU (1,256,40,40),(128,256,1,1),(128) (1,128,40,40) 0 0 0 0 \ 2 2112 2176 633.00 Conv:Conv_37
D RKNN: [23:18:44.618] 20 ConvRelu INT8 NPU (1,128,40,40),(128,128,1,1),(128) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 417.00 Conv:Conv_39
D RKNN: [23:18:44.618] 21 ConvReluAdd INT8 NPU (1,128,40,40),(128,128,3,3),(128),(1,128,40,40) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 745.00 Conv:Conv_41
D RKNN: [23:18:44.618] 22 ConvRelu INT8 NPU (1,128,40,40),(128,128,1,1),(128) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 417.00 Conv:Conv_44
D RKNN: [23:18:44.618] 23 ConvReluAdd INT8 NPU (1,128,40,40),(128,128,3,3),(128),(1,128,40,40) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 745.00 Conv:Conv_46
D RKNN: [23:18:44.618] 24 ConvRelu INT8 NPU (1,128,40,40),(128,128,1,1),(128) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 417.00 Conv:Conv_49
D RKNN: [23:18:44.618] 25 ConvReluAdd INT8 NPU (1,128,40,40),(128,128,3,3),(128),(1,128,40,40) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 745.00 Conv:Conv_51
D RKNN: [23:18:44.618] 26 ConvRelu INT8 NPU (1,256,40,40),(128,256,1,1),(128) (1,128,40,40) 0 0 0 0 \ 2 2112 2176 633.00 Conv:Conv_54
D RKNN: [23:18:44.618] 27 Concat INT8 NPU (1,128,40,40),(1,128,40,40) (1,256,40,40) 0 0 0 0 \ 2 1104 1280 800.00 Concat:Concat_56
D RKNN: [23:18:44.618] 28 ConvRelu INT8 NPU (1,256,40,40),(256,256,1,1),(256) (1,256,40,40) 0 0 0 0 \ 4 2544 2688 866.00 Conv:Conv_57
D RKNN: [23:18:44.618] 29 ConvRelu INT8 NPU (1,256,40,40),(512,256,3,3),(512) (1,512,20,20) 0 0 0 0 \ 3 2336 2432 1756.00 Conv:Conv_59
D RKNN: [23:18:44.618] 30 ConvRelu INT8 NPU (1,512,20,20),(256,512,1,1),(256) (1,256,20,20) 0 0 0 0 \ 4 2464 2688 430.00 Conv:Conv_61
D RKNN: [23:18:44.618] 31 ConvRelu INT8 NPU (1,256,20,20),(256,256,1,1),(256) (1,256,20,20) 0 0 0 0 \ 1 1056 1088 266.00 Conv:Conv_63
D RKNN: [23:18:44.618] 32 ConvReluAdd INT8 NPU (1,256,20,20),(256,256,3,3),(256),(1,256,20,20) (1,256,20,20) 0 0 0 0 \ 1 1056 1088 878.00 Conv:Conv_65
D RKNN: [23:18:44.618] 33 ConvRelu INT8 NPU (1,512,20,20),(256,512,1,1),(256) (1,256,20,20) 0 0 0 0 \ 4 2464 2688 430.00 Conv:Conv_68
D RKNN: [23:18:44.618] 34 Concat INT8 NPU (1,256,20,20),(1,256,20,20) (1,512,20,20) 0 0 0 0 \ 2 1104 1280 400.00 Concat:Concat_70
D RKNN: [23:18:44.618] 35 ConvRelu INT8 NPU (1,512,20,20),(512,512,1,1),(512) (1,512,20,20) 0 0 0 0 \ 8 3168 3712 660.00 Conv:Conv_71
D RKNN: [23:18:44.618] 36 ConvRelu INT8 NPU (1,512,20,20),(256,512,1,1),(256) (1,256,20,20) 0 0 0 0 \ 4 2464 2688 430.00 Conv:Conv_73
D RKNN: [23:18:44.618] 37 MaxPool INT8 NPU (1,256,20,20) (1,256,20,20) 0 0 0 0 \ 2 464 640 200.00 MaxPool:MaxPool_75
D RKNN: [23:18:44.618] 38 MaxPool INT8 NPU (1,256,20,20) (1,256,20,20) 0 0 0 0 \ 2 464 640 200.00 MaxPool:MaxPool_76
D RKNN: [23:18:44.618] 39 MaxPool INT8 NPU (1,256,20,20) (1,256,20,20) 0 0 0 0 \ 2 464 640 200.00 MaxPool:MaxPool_77
D RKNN: [23:18:44.618] 40 Concat INT8 NPU (1,256,20,20),(1,256,20,20),(1,256,20,20),(1,256,20,20) (1,1024,20,20) 0 0 0 0 \ 4 1216 1536 800.00 Concat:Concat_78
D RKNN: [23:18:44.618] 41 ConvRelu INT8 NPU (1,1024,20,20),(512,1024,1,1),(512) (1,512,20,20) 0 0 0 0 \ 32 7472 9856 1116.00 Conv:Conv_79
D RKNN: [23:18:44.618] 42 ConvRelu INT8 NPU (1,512,20,20),(256,512,1,1),(256) (1,256,20,20) 0 0 0 0 \ 4 2464 2688 430.00 Conv:Conv_81
D RKNN: [23:18:44.618] 43 ConvTranspose INT8 NPU (1,256,20,20),(256,1,4,4),(256) (1,256,40,40) 0 0 0 0 \ 1 1056 1088 506.00 ConvTranspose:Resize_84_2deconv
D RKNN: [23:18:44.618] 44 Concat INT8 NPU (1,256,40,40),(1,256,40,40) (1,512,40,40) 0 0 0 0 \ 2 1104 1280 1600.00 Concat:Concat_85
D RKNN: [23:18:44.618] 45 ConvRelu INT8 NPU (1,512,40,40),(128,512,1,1),(128) (1,128,40,40) 0 0 0 0 \ 8 3248 3712 1065.00 Conv:Conv_86
D RKNN: [23:18:44.618] 46 ConvRelu INT8 NPU (1,128,40,40),(128,128,1,1),(128) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 417.00 Conv:Conv_88
D RKNN: [23:18:44.618] 47 ConvRelu INT8 NPU (1,128,40,40),(128,128,3,3),(128) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 545.00 Conv:Conv_90
D RKNN: [23:18:44.618] 48 ConvRelu INT8 NPU (1,512,40,40),(128,512,1,1),(128) (1,128,40,40) 0 0 0 0 \ 8 3248 3712 1065.00 Conv:Conv_92
D RKNN: [23:18:44.618] 49 Concat INT8 NPU (1,128,40,40),(1,128,40,40) (1,256,40,40) 0 0 0 0 \ 2 1104 1280 800.00 Concat:Concat_94
D RKNN: [23:18:44.618] 50 ConvRelu INT8 NPU (1,256,40,40),(256,256,1,1),(256) (1,256,40,40) 0 0 0 0 \ 4 2544 2688 866.00 Conv:Conv_95
D RKNN: [23:18:44.618] 51 ConvRelu INT8 NPU (1,256,40,40),(128,256,1,1),(128) (1,128,40,40) 0 0 0 0 \ 2 2112 2176 633.00 Conv:Conv_97
D RKNN: [23:18:44.618] 52 ConvTranspose INT8 NPU (1,128,40,40),(128,1,4,4),(128) (1,128,80,80) 0 0 0 0 \ 1 1056 1088 1003.00 ConvTranspose:Resize_100_2deconv
D RKNN: [23:18:44.618] 53 Concat INT8 NPU (1,128,80,80),(1,128,80,80) (1,256,80,80) 0 0 0 0 \ 2 1104 1280 3200.00 Concat:Concat_101
D RKNN: [23:18:44.618] 54 ConvRelu INT8 NPU (1,256,80,80),(64,256,1,1),(64) (1,64,80,80) 0 0 0 0 \ 8 3208 3712 2016.50 Conv:Conv_102
D RKNN: [23:18:44.618] 55 ConvRelu INT8 NPU (1,64,80,80),(64,64,1,1),(64) (1,64,80,80) 0 0 0 0 \ 2 2112 2176 804.50 Conv:Conv_104
D RKNN: [23:18:44.618] 56 ConvRelu INT8 NPU (1,64,80,80),(64,64,3,3),(64) (1,64,80,80) 0 0 0 0 \ 2 2112 2176 836.50 Conv:Conv_106
D RKNN: [23:18:44.618] 57 ConvRelu INT8 NPU (1,256,80,80),(64,256,1,1),(64) (1,64,80,80) 0 0 0 0 \ 8 3208 3712 2016.50 Conv:Conv_108
D RKNN: [23:18:44.618] 58 Concat INT8 NPU (1,64,80,80),(1,64,80,80) (1,128,80,80) 0 0 0 0 \ 2 1104 1280 1600.00 Concat:Concat_110
D RKNN: [23:18:44.618] 59 ConvRelu INT8 NPU (1,128,80,80),(128,128,1,1),(128) (1,128,80,80) 0 0 0 0 \ 4 2504 2688 1617.00 Conv:Conv_111
D RKNN: [23:18:44.618] 60 Conv INT8 NPU (1,128,80,80),(255,128,1,1),(255) (1,255,80,80) 0 0 0 0 \ 4 2504 2688 2433.88 Conv:Conv_141
D RKNN: [23:18:44.618] 61 OutputOperator INT8 CPU (1,255,80,80) \ 0 0 0 0 \ 0 0 0 1600.00 OutputOperator:269
D RKNN: [23:18:44.618] 62 ConvRelu INT8 NPU (1,128,80,80),(128,128,3,3),(128) (1,128,40,40) 0 0 0 0 \ 4 2552 2688 1145.00 Conv:Conv_113
D RKNN: [23:18:44.618] 63 Concat INT8 NPU (1,128,40,40),(1,128,40,40) (1,256,40,40) 0 0 0 0 \ 2 1104 1280 800.00 Concat:Concat_115
D RKNN: [23:18:44.618] 64 ConvRelu INT8 NPU (1,256,40,40),(128,256,1,1),(128) (1,128,40,40) 0 0 0 0 \ 2 2112 2176 633.00 Conv:Conv_116
D RKNN: [23:18:44.618] 65 ConvRelu INT8 NPU (1,128,40,40),(128,128,1,1),(128) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 417.00 Conv:Conv_118
D RKNN: [23:18:44.618] 66 ConvRelu INT8 NPU (1,128,40,40),(128,128,3,3),(128) (1,128,40,40) 0 0 0 0 \ 1 1056 1088 545.00 Conv:Conv_120
D RKNN: [23:18:44.618] 67 ConvRelu INT8 NPU (1,256,40,40),(128,256,1,1),(128) (1,128,40,40) 0 0 0 0 \ 2 2112 2176 633.00 Conv:Conv_122
D RKNN: [23:18:44.618] 68 Concat INT8 NPU (1,128,40,40),(1,128,40,40) (1,256,40,40) 0 0 0 0 \ 2 1104 1280 800.00 Concat:Concat_124
D RKNN: [23:18:44.618] 69 ConvRelu INT8 NPU (1,256,40,40),(256,256,1,1),(256) (1,256,40,40) 0 0 0 0 \ 4 2544 2688 866.00 Conv:Conv_125
D RKNN: [23:18:44.618] 70 Conv INT8 NPU (1,256,40,40),(255,256,1,1),(255) (1,255,40,40) 0 0 0 0 \ 4 2544 2688 865.75 Conv:Conv_143
D RKNN: [23:18:44.618] 71 OutputOperator INT8 CPU (1,255,40,40) \ 0 0 0 0 \ 0 0 0 400.00 OutputOperator:271
D RKNN: [23:18:44.618] 72 ConvRelu INT8 NPU (1,256,40,40),(256,256,3,3),(256) (1,256,20,20) 0 0 0 0 \ 3 2336 2432 1078.00 Conv:Conv_127
D RKNN: [23:18:44.618] 73 Concat INT8 NPU (1,256,20,20),(1,256,20,20) (1,512,20,20) 0 0 0 0 \ 2 1104 1280 400.00 Concat:Concat_129
D RKNN: [23:18:44.618] 74 ConvRelu INT8 NPU (1,512,20,20),(256,512,1,1),(256) (1,256,20,20) 0 0 0 0 \ 4 2464 2688 430.00 Conv:Conv_130
D RKNN: [23:18:44.618] 75 ConvRelu INT8 NPU (1,256,20,20),(256,256,1,1),(256) (1,256,20,20) 0 0 0 0 \ 1 1056 1088 266.00 Conv:Conv_132
D RKNN: [23:18:44.618] 76 ConvRelu INT8 NPU (1,256,20,20),(256,256,3,3),(256) (1,256,20,20) 0 0 0 0 \ 1 1056 1088 778.00 Conv:Conv_134
D RKNN: [23:18:44.618] 77 ConvRelu INT8 NPU (1,512,20,20),(256,512,1,1),(256) (1,256,20,20) 0 0 0 0 \ 4 2464 2688 430.00 Conv:Conv_136
D RKNN: [23:18:44.618] 78 Concat INT8 NPU (1,256,20,20),(1,256,20,20) (1,512,20,20) 0 0 0 0 \ 2 1104 1280 400.00 Concat:Concat_138
D RKNN: [23:18:44.618] 79 ConvRelu INT8 NPU (1,512,20,20),(512,512,1,1),(512) (1,512,20,20) 0 0 0 0 \ 8 3168 3712 660.00 Conv:Conv_139
D RKNN: [23:18:44.618] 80 Conv INT8 NPU (1,512,20,20),(255,512,1,1),(255) (1,255,20,20) 0 0 0 0 \ 4 2488 2688 429.50 Conv:Conv_145
D RKNN: [23:18:44.618] 81 OutputOperator INT8 CPU (1,255,20,20) \ 0 0 0 0 \ 0 0 0 100.00 OutputOperator:273
D RKNN: [23:18:44.618] <<<<<<<< end: N4rknn18RKNNModelBuildPassE
D RKNN: [23:18:44.618] >>>>>> start: N4rknn21RKNNMemStatisticsPassE
D RKNN: [23:18:44.619] ---------------------------------------------------------------------+---------------------------------
D RKNN: [23:18:44.619] ID User Tensor DataType OrigShape NativeShape | [Start End) Size
D RKNN: [23:18:44.619] ---------------------------------------------------------------------+---------------------------------
D RKNN: [23:18:44.619] 1 ConvRelu images INT8 (1,3,640,640) (1,1,640,640,3) | 0x00000000 0x0012c000 0x0012c000
D RKNN: [23:18:44.619] 2 ConvRelu 123 INT8 (1,32,320,320) (1,4,320,320,8) | 0x0012c000 0x0044c000 0x00320000
D RKNN: [23:18:44.619] 3 ConvRelu 125 INT8 (1,64,160,160) (1,8,160,160,8) | 0x0044c000*0x005dc000 0x00190000
D RKNN: [23:18:44.619] 4 ConvRelu 127 INT8 (1,32,160,160) (1,4,160,160,8) | 0x00000000 0x000c8000 0x000c8000
D RKNN: [23:18:44.619] 5 ConvReluAdd 129 INT8 (1,32,160,160) (1,4,160,160,8) | 0x000c8000 0x00190000 0x000c8000
D RKNN: [23:18:44.619] 5 ConvReluAdd 127 INT8 (1,32,160,160) (1,4,160,160,8) | 0x00000000 0x000c8000 0x000c8000
D RKNN: [23:18:44.619] 6 ConvRelu 125 INT8 (1,64,160,160) (1,8,160,160,8) | 0x0044c000*0x005dc000 0x00190000
D RKNN: [23:18:44.619] 7 Concat 132 INT8 (1,32,160,160) (1,4,160,160,8) | 0x00190000 0x00258000 0x000c8000
D RKNN: [23:18:44.619] 7 Concat 134 INT8 (1,32,160,160) (1,4,160,160,8) | 0x00000000 0x000c8000 0x000c8000
D RKNN: [23:18:44.619] 8 ConvRelu 135 INT8 (1,64,160,160) (1,8,160,160,8) | 0x00258000 0x003e8000 0x00190000
D RKNN: [23:18:44.619] 9 ConvRelu 137 INT8 (1,64,160,160) (1,8,160,160,8) | 0x00000000 0x00190000 0x00190000
D RKNN: [23:18:44.619] 10 ConvRelu 139 INT8 (1,128,80,80) (1,16,80,80,8) | 0x00190000 0x00258000 0x000c8000
D RKNN: [23:18:44.619] 11 ConvRelu 141 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00000000 0x00064000 0x00064000
D RKNN: [23:18:44.619] 12 ConvReluAdd 143 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00064000 0x000c8000 0x00064000
D RKNN: [23:18:44.619] 12 ConvReluAdd 141 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00000000 0x00064000 0x00064000
D RKNN: [23:18:44.619] 13 ConvRelu 146 INT8 (1,64,80,80) (1,8,80,80,8) | 0x000c8000 0x0012c000 0x00064000
D RKNN: [23:18:44.619] 14 ConvReluAdd 148 INT8 (1,64,80,80) (1,8,80,80,8) | 0x0012c000 0x00190000 0x00064000
D RKNN: [23:18:44.619] 14 ConvReluAdd 146 INT8 (1,64,80,80) (1,8,80,80,8) | 0x000c8000 0x0012c000 0x00064000
D RKNN: [23:18:44.619] 15 ConvRelu 139 INT8 (1,128,80,80) (1,16,80,80,8) | 0x00190000 0x00258000 0x000c8000
D RKNN: [23:18:44.619] 16 Concat 151 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00000000 0x00064000 0x00064000
D RKNN: [23:18:44.619] 16 Concat 153 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00064000 0x000c8000 0x00064000
D RKNN: [23:18:44.619] 17 ConvRelu 154 INT8 (1,128,80,80) (1,16,80,80,8) | 0x000c8000 0x00190000 0x000c8000
D RKNN: [23:18:44.619] 18 ConvRelu 156 INT8 (1,128,80,80) (1,16,80,80,8) | 0x00000000 0x000c8000 0x000c8000
D RKNN: [23:18:44.619] 19 ConvRelu 158 INT8 (1,256,40,40) (1,32,40,40,8) | 0x000c8000 0x0012c000 0x00064000
D RKNN: [23:18:44.619] 20 ConvRelu 160 INT8 (1,128,40,40) (1,16,40,40,8) | 0x0012c000 0x0015e000 0x00032000
D RKNN: [23:18:44.619] 21 ConvReluAdd 162 INT8 (1,128,40,40) (1,16,40,40,8) | 0x0015e000 0x00190000 0x00032000
D RKNN: [23:18:44.619] 21 ConvReluAdd 160 INT8 (1,128,40,40) (1,16,40,40,8) | 0x0012c000 0x0015e000 0x00032000
D RKNN: [23:18:44.619] 22 ConvRelu 165 INT8 (1,128,40,40) (1,16,40,40,8) | 0x00190000 0x001c2000 0x00032000
D RKNN: [23:18:44.619] 23 ConvReluAdd 167 INT8 (1,128,40,40) (1,16,40,40,8) | 0x0012c000 0x0015e000 0x00032000
D RKNN: [23:18:44.619] 23 ConvReluAdd 165 INT8 (1,128,40,40) (1,16,40,40,8) | 0x00190000 0x001c2000 0x00032000
D RKNN: [23:18:44.619] 24 ConvRelu 170 INT8 (1,128,40,40) (1,16,40,40,8) | 0x0015e000 0x00190000 0x00032000
D RKNN: [23:18:44.619] 25 ConvReluAdd 172 INT8 (1,128,40,40) (1,16,40,40,8) | 0x0012c000 0x0015e000 0x00032000
D RKNN: [23:18:44.619] 25 ConvReluAdd 170 INT8 (1,128,40,40) (1,16,40,40,8) | 0x0015e000 0x00190000 0x00032000
D RKNN: [23:18:44.619] 26 ConvRelu 158 INT8 (1,256,40,40) (1,32,40,40,8) | 0x000c8000 0x0012c000 0x00064000
D RKNN: [23:18:44.619] 27 Concat 175 INT8 (1,128,40,40) (1,16,40,40,8) | 0x00190000 0x001c2000 0x00032000
D RKNN: [23:18:44.619] 27 Concat 177 INT8 (1,128,40,40) (1,16,40,40,8) | 0x0012c000 0x0015e000 0x00032000
D RKNN: [23:18:44.619] 28 ConvRelu 178 INT8 (1,256,40,40) (1,32,40,40,8) | 0x000c8000 0x0012c000 0x00064000
D RKNN: [23:18:44.619] 29 ConvRelu 180 INT8 (1,256,40,40) (1,35,40,40,8) | 0x0012c000 0x00199920 0x0006d920
D RKNN: [23:18:44.619] 30 ConvRelu 182 INT8 (1,512,20,20) (1,64,20,20,8) | 0x000c8000 0x000fa000 0x00032000
D RKNN: [23:18:44.619] 31 ConvRelu 184 INT8 (1,256,20,20) (1,35,20,20,8) | 0x000fa000 0x00115648 0x0001b648
D RKNN: [23:18:44.619] 32 ConvReluAdd 186 INT8 (1,256,20,20) (1,32,20,20,8) | 0x00196400 0x001af400 0x00019000
D RKNN: [23:18:44.619] 32 ConvReluAdd 184 INT8 (1,256,20,20) (1,35,20,20,8) | 0x000fa000 0x00115648 0x0001b648
D RKNN: [23:18:44.619] 33 ConvRelu 182 INT8 (1,512,20,20) (1,64,20,20,8) | 0x000c8000 0x000fa000 0x00032000
D RKNN: [23:18:44.619] 34 Concat 189 INT8 (1,256,20,20) (1,32,20,20,8) | 0x001af400 0x001c8400 0x00019000
D RKNN: [23:18:44.619] 34 Concat 191 INT8 (1,256,20,20) (1,35,20,20,8) | 0x000fa000 0x00115648 0x0001b648
D RKNN: [23:18:44.619] 35 ConvRelu 192 INT8 (1,512,20,20) (1,64,20,20,8) | 0x000c8000 0x000fa000 0x00032000
D RKNN: [23:18:44.619] 36 ConvRelu 194 INT8 (1,512,20,20) (1,67,20,20,8) | 0x00196400 0x001ca9e0 0x000345e0
D RKNN: [23:18:44.619] 37 MaxPool 196 INT8 (1,256,20,20) (1,35,20,20,8) | 0x000c8000 0x000e3648 0x0001b648
D RKNN: [23:18:44.619] 38 MaxPool 197 INT8 (1,256,20,20) (1,32,20,20,8) | 0x000e2900 0x000fb900 0x00019000
D RKNN: [23:18:44.619] 39 MaxPool 198 INT8 (1,256,20,20) (1,32,20,20,8) | 0x000fb900 0x00114900 0x00019000
D RKNN: [23:18:44.619] 40 Concat 196 INT8 (1,256,20,20) (1,35,20,20,8) | 0x000c8000 0x000e3648 0x0001b648
D RKNN: [23:18:44.619] 40 Concat 197 INT8 (1,256,20,20) (1,32,20,20,8) | 0x000e2900 0x000fb900 0x00019000
D RKNN: [23:18:44.619] 40 Concat 198 INT8 (1,256,20,20) (1,32,20,20,8) | 0x000fb900 0x00114900 0x00019000
D RKNN: [23:18:44.619] 40 Concat 199 INT8 (1,256,20,20) (1,32,20,20,8) | 0x00196400 0x001af400 0x00019000
D RKNN: [23:18:44.619] 41 ConvRelu 200 INT8 (1,1024,20,20) (1,128,20,20,8) | 0x001af400 0x00213400 0x00064000
D RKNN: [23:18:44.619] 42 ConvRelu 202 INT8 (1,512,20,20) (1,67,20,20,8) | 0x000c8000 0x000fc5e0 0x000345e0
D RKNN: [23:18:44.619] 43 ConvTranspose 204 INT8 (1,256,20,20) (1,35,20,20,8) | 0x000fb900 0x00116f48 0x0001b648
D RKNN: [23:18:44.619] 44 Concat 209 INT8 (1,256,40,40) (1,32,40,40,8) | 0x00196400 0x001fa400 0x00064000
D RKNN: [23:18:44.619] 44 Concat 180 INT8 (1,256,40,40) (1,35,40,40,8) | 0x0012c000 0x00199920 0x0006d920
D RKNN: [23:18:44.619] 45 ConvRelu 210 INT8 (1,512,40,40) (1,64,40,40,8) | 0x001fa400 0x002c2400 0x000c8000
D RKNN: [23:18:44.619] 46 ConvRelu 212 INT8 (1,128,40,40) (1,19,40,40,8) | 0x00116200 0x00151e40 0x0003bc40
D RKNN: [23:18:44.619] 47 ConvRelu 214 INT8 (1,128,40,40) (1,16,40,40,8) | 0x000c8000 0x000fa000 0x00032000
D RKNN: [23:18:44.619] 48 ConvRelu 210 INT8 (1,512,40,40) (1,64,40,40,8) | 0x001fa400 0x002c2400 0x000c8000
D RKNN: [23:18:44.619] 49 Concat 216 INT8 (1,128,40,40) (1,16,40,40,8) | 0x00116200 0x00148200 0x00032000
D RKNN: [23:18:44.619] 49 Concat 218 INT8 (1,128,40,40) (1,19,40,40,8) | 0x00148200 0x00183e40 0x0003bc40
D RKNN: [23:18:44.619] 50 ConvRelu 219 INT8 (1,256,40,40) (1,32,40,40,8) | 0x00180600 0x001e4600 0x00064000
D RKNN: [23:18:44.619] 51 ConvRelu 221 INT8 (1,256,40,40) (1,35,40,40,8) | 0x00116200 0x00183b20 0x0006d920
D RKNN: [23:18:44.619] 52 ConvTranspose 223 INT8 (1,128,40,40) (1,16,40,40,8) | 0x000c8000 0x000fa000 0x00032000
D RKNN: [23:18:44.619] 53 Concat 228 INT8 (1,128,80,80) (1,16,80,80,8) | 0x00116200 0x001de200 0x000c8000
D RKNN: [23:18:44.619] 53 Concat 156 INT8 (1,128,80,80) (1,16,80,80,8) | 0x00000000 0x000c8000 0x000c8000
D RKNN: [23:18:44.619] 54 ConvRelu 229 INT8 (1,256,80,80) (1,32,80,80,8) | 0x001de200 0x0036e200 0x00190000
D RKNN: [23:18:44.619] 55 ConvRelu 231 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00000000 0x00064000 0x00064000
D RKNN: [23:18:44.619] 56 ConvRelu 233 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00064000 0x000c8000 0x00064000
D RKNN: [23:18:44.619] 57 ConvRelu 229 INT8 (1,256,80,80) (1,32,80,80,8) | 0x001de200 0x0036e200 0x00190000
D RKNN: [23:18:44.619] 58 Concat 235 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00000000 0x00064000 0x00064000
D RKNN: [23:18:44.619] 58 Concat 237 INT8 (1,64,80,80) (1,8,80,80,8) | 0x00064000 0x000c8000 0x00064000
D RKNN: [23:18:44.619] 59 ConvRelu 238 INT8 (1,128,80,80) (1,16,80,80,8) | 0x00116200 0x001de200 0x000c8000
D RKNN: [23:18:44.619] 60 Conv 240 INT8 (1,128,80,80) (1,16,80,80,8) | 0x00000000 0x000c8000 0x000c8000
D RKNN: [23:18:44.619] 61 OutputOperator 269 INT8 (1,255,80,80) (1,32,80,80,8) | 0x00116200 0x002a6200 0x00190000
D RKNN: [23:18:44.619] 62 ConvRelu 240 INT8 (1,128,80,80) (1,16,80,80,8) | 0x00000000 0x000c8000 0x000c8000
D RKNN: [23:18:44.619] 63 Concat 242 INT8 (1,128,40,40) (1,16,40,40,8) | 0x002a6200 0x002d8200 0x00032000
D RKNN: [23:18:44.619] 63 Concat 223 INT8 (1,128,40,40) (1,16,40,40,8) | 0x000c8000 0x000fa000 0x00032000
D RKNN: [23:18:44.619] 64 ConvRelu 243 INT8 (1,256,40,40) (1,32,40,40,8) | 0x00000000 0x00064000 0x00064000
D RKNN: [23:18:44.619] 65 ConvRelu 245 INT8 (1,128,40,40) (1,16,40,40,8) | 0x00064000 0x00096000 0x00032000
D RKNN: [23:18:44.619] 66 ConvRelu 247 INT8 (1,128,40,40) (1,16,40,40,8) | 0x00096000 0x000c8000 0x00032000
D RKNN: [23:18:44.619] 67 ConvRelu 243 INT8 (1,256,40,40) (1,32,40,40,8) | 0x00000000 0x00064000 0x00064000
D RKNN: [23:18:44.619] 68 Concat 249 INT8 (1,128,40,40) (1,16,40,40,8) | 0x00064000 0x00096000 0x00032000
D RKNN: [23:18:44.619] 68 Concat 251 INT8 (1,128,40,40) (1,16,40,40,8) | 0x00096000 0x000c8000 0x00032000
D RKNN: [23:18:44.619] 69 ConvRelu 252 INT8 (1,256,40,40) (1,32,40,40,8) | 0x00000000 0x00064000 0x00064000
D RKNN: [23:18:44.619] 70 Conv 254 INT8 (1,256,40,40) (1,35,40,40,8) | 0x00064000 0x000d1920 0x0006d920
D RKNN: [23:18:44.619] 71 OutputOperator 271 INT8 (1,255,40,40) (1,35,40,40,8) | 0x002a6200 0x00313b50 0x0006d950
D RKNN: [23:18:44.619] 72 ConvRelu 254 INT8 (1,256,40,40) (1,35,40,40,8) | 0x00064000 0x000d1920 0x0006d920
D RKNN: [23:18:44.619] 73 Concat 256 INT8 (1,256,20,20) (1,32,20,20,8) | 0x000ce400 0x000e7400 0x00019000
D RKNN: [23:18:44.619] 73 Concat 204 INT8 (1,256,20,20) (1,35,20,20,8) | 0x000fb900 0x00116f48 0x0001b648
D RKNN: [23:18:44.619] 74 ConvRelu 257 INT8 (1,512,20,20) (1,64,20,20,8) | 0x00000000 0x00032000 0x00032000
D RKNN: [23:18:44.619] 75 ConvRelu 259 INT8 (1,256,20,20) (1,35,20,20,8) | 0x00032000 0x0004d648 0x0001b648
D RKNN: [23:18:44.619] 76 ConvRelu 261 INT8 (1,256,20,20) (1,32,20,20,8) | 0x0004c900 0x00065900 0x00019000
D RKNN: [23:18:44.619] 77 ConvRelu 257 INT8 (1,512,20,20) (1,64,20,20,8) | 0x00000000 0x00032000 0x00032000
D RKNN: [23:18:44.619] 78 Concat 263 INT8 (1,256,20,20) (1,32,20,20,8) | 0x00032000 0x0004b000 0x00019000
D RKNN: [23:18:44.619] 78 Concat 265 INT8 (1,256,20,20) (1,35,20,20,8) | 0x0004b000 0x00066648 0x0001b648
D RKNN: [23:18:44.619] 79 ConvRelu 266 INT8 (1,512,20,20) (1,64,20,20,8) | 0x00000000 0x00032000 0x00032000
D RKNN: [23:18:44.619] 80 Conv 268 INT8 (1,512,20,20) (1,67,20,20,8) | 0x00032000 0x000665e0 0x000345e0
D RKNN: [23:18:44.619] 81 OutputOperator 273 INT8 (1,255,20,20) (1,35,20,20,8) | 0x00000000 0x0001b650 0x0001b650
D RKNN: [23:18:44.619] ---------------------------------------------------------------------+---------------------------------
D RKNN: [23:18:44.619] ----------------------------------------------------------------------------+---------------------------------
D RKNN: [23:18:44.619] ID User Tensor DataType OrigShape | [Start End) Size
D RKNN: [23:18:44.619] ----------------------------------------------------------------------------+---------------------------------
D RKNN: [23:18:44.619] 1 ConvRelu model.0.conv.weight INT8 (32,3,6,6) | 0x00000000 0x00002400 0x00002400
D RKNN: [23:18:44.619] 1 ConvRelu model.0.conv.bias INT32 (32) | 0x00002400 0x00002500 0x00000100
D RKNN: [23:18:44.619] 2 ConvRelu model.1.conv.weight INT8 (64,32,3,3) | 0x00002500 0x00006d00 0x00004800
D RKNN: [23:18:44.619] 2 ConvRelu model.1.conv.bias INT32 (64) | 0x00006d00 0x00006f00 0x00000200
D RKNN: [23:18:44.619] 3 ConvRelu model.2.cv1.conv.weight INT8 (32,64,1,1) | 0x00006f00 0x00007700 0x00000800
D RKNN: [23:18:44.619] 3 ConvRelu model.2.cv1.conv.bias INT32 (32) | 0x00007700 0x00007800 0x00000100
D RKNN: [23:18:44.619] 4 ConvRelu model.2.m.0.cv1.conv.weight INT8 (32,32,1,1) | 0x00007800 0x00007c00 0x00000400
D RKNN: [23:18:44.619] 4 ConvRelu model.2.m.0.cv1.conv.bias INT32 (32) | 0x00007c00 0x00007d00 0x00000100
D RKNN: [23:18:44.619] 5 ConvReluAdd model.2.m.0.cv2.conv.weight INT8 (32,32,3,3) | 0x00007d00 0x0000a100 0x00002400
D RKNN: [23:18:44.619] 5 ConvReluAdd model.2.m.0.cv2.conv.bias INT32 (32) | 0x0000a100 0x0000a200 0x00000100
D RKNN: [23:18:44.619] 6 ConvRelu model.2.cv2.conv.weight INT8 (32,64,1,1) | 0x0000a200 0x0000aa00 0x00000800
D RKNN: [23:18:44.619] 6 ConvRelu model.2.cv2.conv.bias INT32 (32) | 0x0000aa00 0x0000ab00 0x00000100
D RKNN: [23:18:44.619] 8 ConvRelu model.2.cv3.conv.weight INT8 (64,64,1,1) | 0x0000ab00 0x0000bb00 0x00001000
D RKNN: [23:18:44.619] 8 ConvRelu model.2.cv3.conv.bias INT32 (64) | 0x0000bb00 0x0000bd00 0x00000200
D RKNN: [23:18:44.619] 9 ConvRelu model.3.conv.weight INT8 (128,64,3,3) | 0x0000bd00 0x0001dd00 0x00012000
D RKNN: [23:18:44.619] 9 ConvRelu model.3.conv.bias INT32 (128) | 0x0001dd00 0x0001e100 0x00000400
D RKNN: [23:18:44.619] 10 ConvRelu model.4.cv1.conv.weight INT8 (64,128,1,1) | 0x0001e100 0x00020100 0x00002000
D RKNN: [23:18:44.619] 10 ConvRelu model.4.cv1.conv.bias INT32 (64) | 0x00020100 0x00020300 0x00000200
D RKNN: [23:18:44.619] 11 ConvRelu model.4.m.0.cv1.conv.weight INT8 (64,64,1,1) | 0x00020300 0x00021300 0x00001000
D RKNN: [23:18:44.619] 11 ConvRelu model.4.m.0.cv1.conv.bias INT32 (64) | 0x00021300 0x00021500 0x00000200
D RKNN: [23:18:44.619] 12 ConvReluAdd model.4.m.0.cv2.conv.weight INT8 (64,64,3,3) | 0x00021500 0x0002a500 0x00009000
D RKNN: [23:18:44.619] 12 ConvReluAdd model.4.m.0.cv2.conv.bias INT32 (64) | 0x0002a500 0x0002a700 0x00000200
D RKNN: [23:18:44.619] 13 ConvRelu model.4.m.1.cv1.conv.weight INT8 (64,64,1,1) | 0x0002a700 0x0002b700 0x00001000
D RKNN: [23:18:44.619] 13 ConvRelu model.4.m.1.cv1.conv.bias INT32 (64) | 0x0002b700 0x0002b900 0x00000200
D RKNN: [23:18:44.619] 14 ConvReluAdd model.4.m.1.cv2.conv.weight INT8 (64,64,3,3) | 0x0002b900 0x00034900 0x00009000
D RKNN: [23:18:44.619] 14 ConvReluAdd model.4.m.1.cv2.conv.bias INT32 (64) | 0x00034900 0x00034b00 0x00000200
D RKNN: [23:18:44.619] 15 ConvRelu model.4.cv2.conv.weight INT8 (64,128,1,1) | 0x00034b00 0x00036b00 0x00002000
D RKNN: [23:18:44.619] 15 ConvRelu model.4.cv2.conv.bias INT32 (64) | 0x00036b00 0x00036d00 0x00000200
D RKNN: [23:18:44.619] 17 ConvRelu model.4.cv3.conv.weight INT8 (128,128,1,1) | 0x00036d00 0x0003ad00 0x00004000
D RKNN: [23:18:44.619] 17 ConvRelu model.4.cv3.conv.bias INT32 (128) | 0x0003ad00 0x0003b100 0x00000400
D RKNN: [23:18:44.619] 18 ConvRelu model.5.conv.weight INT8 (256,128,3,3) | 0x0003b100 0x00083100 0x00048000
D RKNN: [23:18:44.619] 18 ConvRelu model.5.conv.bias INT32 (256) | 0x00083100 0x00083900 0x00000800
D RKNN: [23:18:44.619] 19 ConvRelu model.6.cv1.conv.weight INT8 (128,256,1,1) | 0x00083900 0x0008b900 0x00008000
D RKNN: [23:18:44.619] 19 ConvRelu model.6.cv1.conv.bias INT32 (128) | 0x0008b900 0x0008bd00 0x00000400
D RKNN: [23:18:44.619] 20 ConvRelu model.6.m.0.cv1.conv.weight INT8 (128,128,1,1) | 0x0008bd00 0x0008fd00 0x00004000
D RKNN: [23:18:44.619] 20 ConvRelu model.6.m.0.cv1.conv.bias INT32 (128) | 0x0008fd00 0x00090100 0x00000400
D RKNN: [23:18:44.619] 21 ConvReluAdd model.6.m.0.cv2.conv.weight INT8 (128,128,3,3) | 0x00090100 0x000b4100 0x00024000
D RKNN: [23:18:44.619] 21 ConvReluAdd model.6.m.0.cv2.conv.bias INT32 (128) | 0x000b4100 0x000b4500 0x00000400
D RKNN: [23:18:44.619] 22 ConvRelu model.6.m.1.cv1.conv.weight INT8 (128,128,1,1) | 0x000b4500 0x000b8500 0x00004000
D RKNN: [23:18:44.619] 22 ConvRelu model.6.m.1.cv1.conv.bias INT32 (128) | 0x000b8500 0x000b8900 0x00000400
D RKNN: [23:18:44.619] 23 ConvReluAdd model.6.m.1.cv2.conv.weight INT8 (128,128,3,3) | 0x000b8900 0x000dc900 0x00024000
D RKNN: [23:18:44.619] 23 ConvReluAdd model.6.m.1.cv2.conv.bias INT32 (128) | 0x000dc900 0x000dcd00 0x00000400
D RKNN: [23:18:44.619] 24 ConvRelu model.6.m.2.cv1.conv.weight INT8 (128,128,1,1) | 0x000dcd00 0x000e0d00 0x00004000
D RKNN: [23:18:44.619] 24 ConvRelu model.6.m.2.cv1.conv.bias INT32 (128) | 0x000e0d00 0x000e1100 0x00000400
D RKNN: [23:18:44.619] 25 ConvReluAdd model.6.m.2.cv2.conv.weight INT8 (128,128,3,3) | 0x000e1100 0x00105100 0x00024000
D RKNN: [23:18:44.619] 25 ConvReluAdd model.6.m.2.cv2.conv.bias INT32 (128) | 0x00105100 0x00105500 0x00000400
D RKNN: [23:18:44.619] 26 ConvRelu model.6.cv2.conv.weight INT8 (128,256,1,1) | 0x00105500 0x0010d500 0x00008000
D RKNN: [23:18:44.619] 26 ConvRelu model.6.cv2.conv.bias INT32 (128) | 0x0010d500 0x0010d900 0x00000400
D RKNN: [23:18:44.619] 28 ConvRelu model.6.cv3.conv.weight INT8 (256,256,1,1) | 0x0010d900 0x0011d900 0x00010000
D RKNN: [23:18:44.619] 28 ConvRelu model.6.cv3.conv.bias INT32 (256) | 0x0011d900 0x0011e100 0x00000800
D RKNN: [23:18:44.619] 29 ConvRelu model.7.conv.weight INT8 (512,256,3,3) | 0x0011e100 0x0023e100 0x00120000
D RKNN: [23:18:44.619] 29 ConvRelu model.7.conv.bias INT32 (512) | 0x0023e100 0x0023f100 0x00001000
D RKNN: [23:18:44.619] 30 ConvRelu model.8.cv1.conv.weight INT8 (256,512,1,1) | 0x0023f100 0x0025f100 0x00020000
D RKNN: [23:18:44.619] 30 ConvRelu model.8.cv1.conv.bias INT32 (256) | 0x0025f100 0x0025f900 0x00000800
D RKNN: [23:18:44.619] 31 ConvRelu model.8.m.0.cv1.conv.weight INT8 (256,256,1,1) | 0x0025f900 0x0026f900 0x00010000
D RKNN: [23:18:44.619] 31 ConvRelu model.8.m.0.cv1.conv.bias INT32 (256) | 0x0026f900 0x00270100 0x00000800
D RKNN: [23:18:44.619] 32 ConvReluAdd model.8.m.0.cv2.conv.weight INT8 (256,256,3,3) | 0x00270100 0x00300100 0x00090000
D RKNN: [23:18:44.619] 32 ConvReluAdd model.8.m.0.cv2.conv.bias INT32 (256) | 0x00300100 0x00300900 0x00000800
D RKNN: [23:18:44.619] 33 ConvRelu model.8.cv2.conv.weight INT8 (256,512,1,1) | 0x00300900 0x00320900 0x00020000
D RKNN: [23:18:44.619] 33 ConvRelu model.8.cv2.conv.bias INT32 (256) | 0x00320900 0x00321100 0x00000800
D RKNN: [23:18:44.619] 35 ConvRelu model.8.cv3.conv.weight INT8 (512,512,1,1) | 0x00321100 0x00361100 0x00040000
D RKNN: [23:18:44.619] 35 ConvRelu model.8.cv3.conv.bias INT32 (512) | 0x00361100 0x00362100 0x00001000
D RKNN: [23:18:44.619] 36 ConvRelu model.9.cv1.conv.weight INT8 (256,512,1,1) | 0x00362100 0x00382100 0x00020000
D RKNN: [23:18:44.619] 36 ConvRelu model.9.cv1.conv.bias INT32 (256) | 0x00382100 0x00382900 0x00000800
D RKNN: [23:18:44.619] 41 ConvRelu model.9.cv2.conv.weight INT8 (512,1024,1,1) | 0x00382900 0x00402900 0x00080000
D RKNN: [23:18:44.619] 41 ConvRelu model.9.cv2.conv.bias INT32 (512) | 0x00402900 0x00403900 0x00001000
D RKNN: [23:18:44.619] 42 ConvRelu model.10.conv.weight INT8 (256,512,1,1) | 0x00403900 0x00423900 0x00020000
D RKNN: [23:18:44.619] 42 ConvRelu model.10.conv.bias INT32 (256) | 0x00423900 0x00424100 0x00000800
D RKNN: [23:18:44.619] 43 ConvTranspose Resize_84_2deconv_weight INT8 (256,1,4,4) | 0x006f7180 0x006f8180 0x00001000
D RKNN: [23:18:44.619] 43 ConvTranspose Resize_84_2deconv_weight_bias_0 INT32 (256) | 0x006f8980 0x006f9180 0x00000800
D RKNN: [23:18:44.619] 45 ConvRelu model.13.cv1.conv.weight INT8 (128,512,1,1) | 0x00424100 0x00434100 0x00010000
D RKNN: [23:18:44.619] 45 ConvRelu model.13.cv1.conv.bias INT32 (128) | 0x00434100 0x00434500 0x00000400
D RKNN: [23:18:44.619] 46 ConvRelu model.13.m.0.cv1.conv.weight INT8 (128,128,1,1) | 0x00434500 0x00438500 0x00004000
D RKNN: [23:18:44.619] 46 ConvRelu model.13.m.0.cv1.conv.bias INT32 (128) | 0x00438500 0x00438900 0x00000400
D RKNN: [23:18:44.619] 47 ConvRelu model.13.m.0.cv2.conv.weight INT8 (128,128,3,3) | 0x00438900 0x0045c900 0x00024000
D RKNN: [23:18:44.619] 47 ConvRelu model.13.m.0.cv2.conv.bias INT32 (128) | 0x0045c900 0x0045cd00 0x00000400
D RKNN: [23:18:44.619] 48 ConvRelu model.13.cv2.conv.weight INT8 (128,512,1,1) | 0x0045cd00 0x0046cd00 0x00010000
D RKNN: [23:18:44.619] 48 ConvRelu model.13.cv2.conv.bias INT32 (128) | 0x0046cd00 0x0046d100 0x00000400
D RKNN: [23:18:44.619] 50 ConvRelu model.13.cv3.conv.weight INT8 (256,256,1,1) | 0x0046d100 0x0047d100 0x00010000
D RKNN: [23:18:44.619] 50 ConvRelu model.13.cv3.conv.bias INT32 (256) | 0x0047d100 0x0047d900 0x00000800
D RKNN: [23:18:44.619] 51 ConvRelu model.14.conv.weight INT8 (128,256,1,1) | 0x0047d900 0x00485900 0x00008000
D RKNN: [23:18:44.619] 51 ConvRelu model.14.conv.bias INT32 (128) | 0x00485900 0x00485d00 0x00000400
D RKNN: [23:18:44.619] 52 ConvTranspose Resize_100_2deconv_weight INT8 (128,1,4,4) | 0x006f8180 0x006f8980 0x00000800
D RKNN: [23:18:44.619] 52 ConvTranspose Resize_100_2deconv_weight_bias_0 INT32 (128) | 0x006f9180*0x006f9580 0x00000400
D RKNN: [23:18:44.619] 54 ConvRelu model.17.cv1.conv.weight INT8 (64,256,1,1) | 0x00485d00 0x00489d00 0x00004000
D RKNN: [23:18:44.619] 54 ConvRelu model.17.cv1.conv.bias INT32 (64) | 0x00489d00 0x00489f00 0x00000200
D RKNN: [23:18:44.619] 55 ConvRelu model.17.m.0.cv1.conv.weight INT8 (64,64,1,1) | 0x00489f00 0x0048af00 0x00001000
D RKNN: [23:18:44.619] 55 ConvRelu model.17.m.0.cv1.conv.bias INT32 (64) | 0x0048af00 0x0048b100 0x00000200
D RKNN: [23:18:44.619] 56 ConvRelu model.17.m.0.cv2.conv.weight INT8 (64,64,3,3) | 0x0048b100 0x00494100 0x00009000
D RKNN: [23:18:44.619] 56 ConvRelu model.17.m.0.cv2.conv.bias INT32 (64) | 0x00494100 0x00494300 0x00000200
D RKNN: [23:18:44.619] 57 ConvRelu model.17.cv2.conv.weight INT8 (64,256,1,1) | 0x00494300 0x00498300 0x00004000
D RKNN: [23:18:44.619] 57 ConvRelu model.17.cv2.conv.bias INT32 (64) | 0x00498300 0x00498500 0x00000200
D RKNN: [23:18:44.619] 59 ConvRelu model.17.cv3.conv.weight INT8 (128,128,1,1) | 0x00498500 0x0049c500 0x00004000
D RKNN: [23:18:44.619] 59 ConvRelu model.17.cv3.conv.bias INT32 (128) | 0x0049c500 0x0049c900 0x00000400
D RKNN: [23:18:44.619] 60 Conv model.24.m.0.weight INT8 (255,128,1,1) | 0x006bdd00 0x006c5c80 0x00007f80
D RKNN: [23:18:44.619] 60 Conv model.24.m.0.bias INT32 (255) | 0x006c5c80 0x006c6480 0x00000800
D RKNN: [23:18:44.619] 62 ConvRelu model.18.conv.weight INT8 (128,128,3,3) | 0x0049c900 0x004c0900 0x00024000
D RKNN: [23:18:44.619] 62 ConvRelu model.18.conv.bias INT32 (128) | 0x004c0900 0x004c0d00 0x00000400
D RKNN: [23:18:44.619] 64 ConvRelu model.20.cv1.conv.weight INT8 (128,256,1,1) | 0x004c0d00 0x004c8d00 0x00008000
D RKNN: [23:18:44.619] 64 ConvRelu model.20.cv1.conv.bias INT32 (128) | 0x004c8d00 0x004c9100 0x00000400
D RKNN: [23:18:44.619] 65 ConvRelu model.20.m.0.cv1.conv.weight INT8 (128,128,1,1) | 0x004c9100 0x004cd100 0x00004000
D RKNN: [23:18:44.619] 65 ConvRelu model.20.m.0.cv1.conv.bias INT32 (128) | 0x004cd100 0x004cd500 0x00000400
D RKNN: [23:18:44.619] 66 ConvRelu model.20.m.0.cv2.conv.weight INT8 (128,128,3,3) | 0x004cd500 0x004f1500 0x00024000
D RKNN: [23:18:44.619] 66 ConvRelu model.20.m.0.cv2.conv.bias INT32 (128) | 0x004f1500 0x004f1900 0x00000400
D RKNN: [23:18:44.619] 67 ConvRelu model.20.cv2.conv.weight INT8 (128,256,1,1) | 0x004f1900 0x004f9900 0x00008000
D RKNN: [23:18:44.619] 67 ConvRelu model.20.cv2.conv.bias INT32 (128) | 0x004f9900 0x004f9d00 0x00000400
D RKNN: [23:18:44.619] 69 ConvRelu model.20.cv3.conv.weight INT8 (256,256,1,1) | 0x004f9d00 0x00509d00 0x00010000
D RKNN: [23:18:44.619] 69 ConvRelu model.20.cv3.conv.bias INT32 (256) | 0x00509d00 0x0050a500 0x00000800
D RKNN: [23:18:44.619] 70 Conv model.24.m.1.weight INT8 (255,256,1,1) | 0x006c6480 0x006d6380 0x0000ff00
D RKNN: [23:18:44.619] 70 Conv model.24.m.1.bias INT32 (255) | 0x006d6380 0x006d6b80 0x00000800
D RKNN: [23:18:44.619] 72 ConvRelu model.21.conv.weight INT8 (256,256,3,3) | 0x0050a500 0x0059a500 0x00090000
D RKNN: [23:18:44.619] 72 ConvRelu model.21.conv.bias INT32 (256) | 0x0059a500 0x0059ad00 0x00000800
D RKNN: [23:18:44.619] 74 ConvRelu model.23.cv1.conv.weight INT8 (256,512,1,1) | 0x0059ad00 0x005bad00 0x00020000
D RKNN: [23:18:44.619] 74 ConvRelu model.23.cv1.conv.bias INT32 (256) | 0x005bad00 0x005bb500 0x00000800
D RKNN: [23:18:44.619] 75 ConvRelu model.23.m.0.cv1.conv.weight INT8 (256,256,1,1) | 0x005bb500 0x005cb500 0x00010000
D RKNN: [23:18:44.619] 75 ConvRelu model.23.m.0.cv1.conv.bias INT32 (256) | 0x005cb500 0x005cbd00 0x00000800
D RKNN: [23:18:44.619] 76 ConvRelu model.23.m.0.cv2.conv.weight INT8 (256,256,3,3) | 0x005cbd00 0x0065bd00 0x00090000
D RKNN: [23:18:44.619] 76 ConvRelu model.23.m.0.cv2.conv.bias INT32 (256) | 0x0065bd00 0x0065c500 0x00000800
D RKNN: [23:18:44.619] 77 ConvRelu model.23.cv2.conv.weight INT8 (256,512,1,1) | 0x0065c500 0x0067c500 0x00020000
D RKNN: [23:18:44.619] 77 ConvRelu model.23.cv2.conv.bias INT32 (256) | 0x0067c500 0x0067cd00 0x00000800
D RKNN: [23:18:44.619] 79 ConvRelu model.23.cv3.conv.weight INT8 (512,512,1,1) | 0x0067cd00 0x006bcd00 0x00040000
D RKNN: [23:18:44.619] 79 ConvRelu model.23.cv3.conv.bias INT32 (512) | 0x006bcd00 0x006bdd00 0x00001000
D RKNN: [23:18:44.619] 80 Conv model.24.m.2.weight INT8 (255,512,1,1) | 0x006d6b80 0x006f6980 0x0001fe00
D RKNN: [23:18:44.619] 80 Conv model.24.m.2.bias INT32 (255) | 0x006f6980 0x006f7180 0x00000800
D RKNN: [23:18:44.619] ----------------------------------------------------------------------------+---------------------------------
D RKNN: [23:18:44.620] ----------------------------------------
D RKNN: [23:18:44.620] Total Weight Memory Size: 7312768
D RKNN: [23:18:44.620] Total Internal Memory Size: 6144000
D RKNN: [23:18:44.620] Predict Internal Memory RW Amount: 82852632
D RKNN: [23:18:44.620] Predict Weight Memory RW Amount: 7312768
D RKNN: [23:18:44.620] ----------------------------------------
D RKNN: [23:18:44.620] <<<<<<<< end: N4rknn21RKNNMemStatisticsPassE
I rknn buiding done.
done
--> Export rknn model
done
--> Init runtime environment
W init_runtime: Target is None, use simulator!
done
--> Running model
W inference: The 'data_format' has not been set and defaults is nhwc!
Analysing : 100%|███████████████████████████████████████████████| 146/146 [00:00<00:00, 6085.53it/s]
Preparing : 100%|████████████████████████████████████████████████| 146/146 [00:01<00:00, 111.14it/s]
W inference: The dims of input(ndarray) shape (640, 640, 3) is wrong, expect dims is 4! Try expand dims to (1, 640, 640, 3)!
done
class: person, score: 0.8838784694671631
box coordinate left,top,right,down: [209.6862335205078, 243.11955797672272, 285.13685607910156, 507.7035621404648]
class: person, score: 0.8669421076774597
box coordinate left,top,right,down: [477.6677174568176, 241.00597953796387, 561.1506419181824, 523.3208637237549]
class: person, score: 0.826057493686676
box coordinate left,top,right,down: [110.24830067157745, 235.76190769672394, 231.76915538311005, 536.1012514829636]
class: person, score: 0.32633307576179504
box coordinate left,top,right,down: [80.75779604911804, 354.98211681842804, 121.49669003486633, 516.531555056572]
class: bus , score: 0.6890695095062256
box coordinate left,top,right,down: [91.16828817129135, 134.78936767578125, 556.8909769654274, 460.78936767578125]
构建成功！！！
验证，不报错则证明安装环境成功！！！
(rknnpy38) ubuntu@ubuntu:~/Downloads/rknn-toolkit2-master/examples/onnx/yolov5$ python
Python 3.8.16 (default, Mar 2 2023, 03:21:46)
[GCC 11.2.0] :: Anaconda, Inc. on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> from rknn.api import RKNN
2 模型转换
将训练好的yolov7.pt转换为yolov7.onnx。
将yolov7.onnx转换为yolov7.rknn。
首先去官网下载源码和yolov7.pt文件。
源码地址：https://github.com/WongKinYiu/yolov7.git
模型地址：https://github.com/WongKinYiu/yolov7/releases/download/v0.1/yolov7.pt
在models/yolo.py中将Detect类中的前向传播过程由：
def forward(self, x):
# x = x.copy() # for profiling
z = [] # inference output
self.training |= self.export
for i in range(self.nl):
x[i] = self.m[i](x[i]) # conv
bs, _, ny, nx = x[i].shape # x(bs,255,20,20) to x(bs,3,20,20,85)
x[i] = x[i].view(bs, self.na, self.no, ny, nx).permute(0, 1, 3, 4, 2).contiguous()
if not self.training: # inference
if self.grid[i].shape[2:4] != x[i].shape[2:4]:
self.grid[i] = self._make_grid(nx, ny).to(x[i].device)
y = x[i].sigmoid()
if not torch.onnx.is_in_onnx_export():
y[..., 0:2] = (y[..., 0:2] * 2. - 0.5 + self.grid[i]) * self.stride[i] # xy
y[..., 2:4] = (y[..., 2:4] * 2) ** 2 * self.anchor_grid[i] # wh
else:
xy, wh, conf = y.split((2, 2, self.nc + 1), 4) # y.tensor_split((2, 4, 5), 4) # torch 1.8.0
xy = xy * (2. * self.stride[i]) + (self.stride[i] * (self.grid[i] - 0.5)) # new xy
wh = wh ** 2 * (4 * self.anchor_grid[i].data) # new wh
y = torch.cat((xy, wh, conf), 4)
z.append(y.view(bs, -1, self.no))
if self.training:
out = x
elif self.end2end:
out = torch.cat(z, 1)
elif self.include_nms:
z = self.convert(z)
out = (z, )
elif self.concat:
out = torch.cat(z, 1)
else:
out = (torch.cat(z, 1), x)
return out
改成：
def forward(self, x):
z = [] # inference output
for i in range(self.nl):
x[i] = self.m[i](x[i])
return x
注：这一步一定要改，否则在onnx转rknn时会报错。
修改后将yolov7.pt移动至export.py同一文件夹下，运行export.py便可以得到yolov7.onnx。
在rknn-toolkit2工程文件夹中浏览至./examples/onnx/yolov5，将我们转换得到的yolov7.onnx复制到该文件夹下，修改该文件夹下的test.py中的内容为自己模型的名字，要修改的地方如下：
此处我们只需要修改前两项和搭载平台的名字即可。
最后运行test.py，即可得到rknn模型。
