rknn-toolkit icon indicating copy to clipboard operation
rknn-toolkit copied to clipboard

Results 212 rknn-toolkit issues
Sort by recently updated
recently updated
newest added

老师好, 目前在做模型部署在RV1126芯片上时遇到问题如下: 编译C++代码时,CMakeLists文件内容如下: `set(CMAKE_CXX_FLAGS_RELEASE "$ENV{CXXFLAGS} -O2 -Wall -Wno-unused-result")` 当将优化等级修改为-O0时,rknn_outputs_get接口返回异常值,无法进行后处理得到结果, 当按照上述设置为-O2时,推理结果正常无异常返回值。 请问具体是什么原因导致rknn_outputs_get接口返回错误值,或可能其他原因导致。

Exception: RKNN init failed. error code: RKNN_ERR_MALLOC_FAIL

![06c907041e2d5c2cb08dd5b70c62078](https://github.com/rockchip-linux/rknn-toolkit/assets/59059785/511eed56-1fda-4fda-b87b-48242c736711)

环境: Linux 639971f8aae7 5.15.0-78-generic #85~20.04.1-Ubuntu SMP Mon Jul 17 09:42:39 UTC 2023 x86_64 x86_64 x86_64 GNU/Linux 版本: rknn-toolkit (1.7.1) 报错信息: [op_check:185]Inputs/Outputs data type not support: INT16, INT32 详请日志信息请查看附件 Exception: RKNN...

**** [log_feedback_to_the_rknn_toolkit_dev_team.log](https://github.com/rockchip-linux/rknn-toolkit/files/12390160/log_feedback_to_the_rknn_toolkit_dev_team.log)

我想转换一个onnx模型,用于时间序列的分类。模型为shufflenet v2,所有conv、gap等都是1d的。输入为[batch, channel, point_num]的npy文件。 load时报警: W Warning: Axis may need to be adjusted according to original model shape. build时报错: E The mean_values length of input 1 must equal 1, such...

class SE(nn.Module): def __init__(self, w=8, c=896, c_tmp=56): super().__init__() self.w=w self.c=c self.c_tmp=c_tmp self.fc1 = nn.Linear(self.c, self.c_tmp) self.relu_func = nn.ReLU(inplace=False) self.fc2 = nn.Linear(self.c_tmp, self.c) self.sigmoid_func = nn.Sigmoid() def forward(self, x): #x =...

测试代码 ` rknn = RKNN() if not os.path.exists(ONNX_MODEL): print('model not exist') exit(-1) # pre-process config print('--> Config model') rknn.config(batch_size=8, quantized_dtype='asymmetric_affine-u8', reorder_channel='0 1 2', mean_values=[[0, 0, 0]], std_values=[[255, 255, 255]], optimization_level=3,...

## 环境版本说明 OS: ubuntu20.04下 docker: docker rknntoolkit 1.7.3 环境 ## 背景: 需要转换Y8的网络,在Q群讨论过了,做了y8的网络层修改, 具体是用原生的y8的网络,做如下 修改 1, 将ultrralytics 文件夹下的所有chunk 算子 等效替换成chunk; 接着便开始输出onnx: ``` from ultralytics import YOLO # Load a model model...

D Real output shape: (2, 4, 4, 1792) I Build torch-jit-export complete. I Clean. MMSE Quant Step 0: 96%|████████████████████████████████████████▏ | 316/330 [00:30