一、写在前面 RKNN-Toolkit2支持的深度学习框架包括Caffe、TensorFlow、TensorFlow Lite、ONNX、DarkNet和PyTorch。 它和各深度学习框架的版本对应关系如下: RKNN-Toolkit2CaffeTensorFlowTF LiteONNXDarkNetPyTo…
模型转换,Toolkit-lite2工具导入原始的Caffe、TensorFlow、TensorFlow Lite、ONNX、Pytorch、MXNet等模型转换成RKNN模型(), 也支持导入RKNN模型然后在NPU平台上加载推理等。 量化功能,支持将浮点模型量化为定点模型,目前支持的量化方法为非对称量化(asymmetric_quantized-8),并支持混合量化功能。 模型推理,能够在PC上模拟...
这是一份原理性的说明,在针对RKNN模型进行int8量化时,在调用pytorch进行建模时,就需要在建模过程中(参数反向传递的过程中,对模型参数进行处理),pytorch也有自己的量化模型,上面这份文档里有很清晰的描述。 还有一份API级别的参考手册,也在github RKNN2.0.0的doc目录,可以在具体编码时查阅。特护注意,一旦建模时出现一...
模型转换:支持 Caffe、TensorFlow、TensorFlow Lite、ONNX、DarkNet、PyTorch 等模型转为 RKNN 模型,并支持 RKNN 模型导入导出,RKNN 模型能够在 Rockchip NPU 平台上加载使用。 量化功能: 支持将浮点模型量化为定点模型 , 目前支持的量化方法为非对称量化(asymmetric_quantized-8),并支持混合量化功能。 模型推理:能够...
models:# model output namename:onnx_detection# Original model frameworkplatform:pytorch# Model input file pathmodel_file_path:./resnet18.pt# Describe information such as input and output shapessubgraphs:# model input tensor shapeinput_size_list:-1,3,512,512# input tensor nameinputs:-data# ou...
Support pytorch 2.1 Improve support for QAT models of pytorch and onnx Optimize automatic generation of C++ code for older version, please refer CHANGELOG Feedback and Community Support Redmine (Feedback recommended, Please consult our sales or FAE for the redmine account) QQ Group Chat: 1025468710...
Support pytorch 2.1 Improve support for QAT models of pytorch and onnx Optimize automatic generation of C++ code for older version, please referCHANGELOG Feedback and Community Support Redmine(Feedback recommended, Please consult our sales or FAE for the redmine account) ...
Improve support for QAT models of pytorch and onnx Optimize automatic generation of C++ code v1.6.0 Support ONNX model of OPSET 12~19 Support custom operators (including CPU and GPU) Improve support for dynamic weight convolution, Layernorm, RoiAlign, Softmax, ReduceL2, Gelu, GLU, etc. Add...
2. support pytorch & mxnet model. 3. support 4 channel model. 4. add loss analysing. 5. add UI for model preview. 6. support setup optimizing level. 7. update hybrid quantization. (more details please see documents.) Version Check: Check correct version before running rknn: RKNNAPI: API...
# Create RKNN object rknn = RKNN() print('--> config model') rknn.config(channel_mean_value='0. 0. 0. 255.', reorder_channel='0 1 2') print('done') # Load pytorch model print('--> Loading model') ret = rknn.load_pytorch(model=model_path, input_size_list=input_size_list)...