kendryte / nncase Goto Github PK

I'm trying to compile a tflite to kmodel with the newest release.
./ncc.exe compile ./mnist.tflite ./mnist.kmodel -i tflite -o kmodel

However got a dataset Error.
Fatal: Invalid dataset, should contain one file at least

Here is the tflite file and the origin keras h5 file:

链接: https://pan.baidu.com/s/1Vd0mUqcjTa-1z4PxYEAe9Q 提取码: ihw6

Is there any plan to release a pre-build binary for Ubuntu? Thanks!

Hi, is there any plan for supporting macOS? Thx

Hello, I encountered this error when converting tiny yolo v2, can you provide the workspace of tiny yolo v2, thanks

Hello, I encountered this error when converting tiny yolo v2, can you provide the workspace of tiny yolo v2, thanks

我将如下模型转换为KMODEL后成功运行，如图中所示，该模型有两个输出层，尺寸分别为21x10x7=1470和21x20x14=5880。但KMODEL显示两个输出层的尺寸分别为5880和23520，请问这是为什么呢？

使用TF尝试构建了一个分割的网络，在PC上运行良好，经过nncase_v0.1转换为kmodel后，主体部分是一堆卷积，和本问题有关的模型的最后几行信息如下：

使用了maix bit开发板，固件0.4.0_44_minimum，下载进上述网络模型kmodel后打印网络信息的最后几行如下：

上述两个网络信息相对比，感觉后者最后的“layer[55] KL_INVALID, 16 bytes”对应前者网络信息的“56：Logistic 1x8x64x128->1x8x64x128”，而这在原网络中是个Sigmoid函数，那么

问题1：加载到开发板后的模型信息显示INVALID等关键字，但Sigmoid函数在nncase_v0.1中是支持的，实际运行中，模型是否真正运行了Sigmoid函数?

由于图像分割最终需要在屏幕上输出一幅图，参考http://blog.sipeed.com/p/673.html#more-673
输出了自己网络第53层的特征图，程序如下：

中间层的fmap可以显示，但是始终无法获得最终的网络输出，加载kmodel运行到不同层，打印出输出的信息如下：

感觉[54]和[55]是参与量化的参数，因此

问题2：如何获得网络最终输出的结果，即最终的经过kmodel中layer[54]和layer[55]后的结果？能生成一幅图像显示在小屏幕上的那种

谢谢！

我用tensorflow官方的MobileNetV2测试，发现并不支持DepthwiseConv2d，是什么原因呢？

Fatal: Layer DepthwiseConv2d is not supported
NnCase.Converter.Converters.LayerNotSupportedException: Layer DepthwiseConv2d is not supported

你好，我想请教你一下yolo.tfilte文件你是怎么训练的？我现在想要训练自己的yolo模型，但是担心训练完的模型，相互转换的时候会出问题。方便的话，你能提供一下代码吗？万分感谢。

是否支持UpSampling2D 和 Deconvolution ？ 另外，pooling是否支持kernel=(3,3)的呢？

0.1版本能正常转换的tflite模型，在0.2中报错不支持tflite中的Logistic操作，请问在nncase中什么操作被认为是Logistic？

Hello, I am using nncase to generate k210model from tflite model. But it requires some images to be present in some folder. But My model is not for image data. It will take reading from other sensors and make decision. Model is a fully connected model(no conv layer) with input shape (1,4).
How do I convert it to kmodel?

HI

I got the following error:

Fatal: Shapes must be same, but got [1x64x56x56] and [1x64x55x55]

liu@EAIU:~/ncc-linux-x86_64$ ncc -i caffe -o k210model --dataset ./images ./mobilenet_yolov3_lite_deploy_iter_3000.caffemodel ./1.kmodel
Fatal: Layer AnnotatedData is not supported
NnCase.Converter.Converters.LayerNotSupportedException: Layer AnnotatedData is not supported
at NnCase.Converter.Converters.CaffeToGraphConverter.ConvertLayer(LayerParameter layerParam) in D:\Work\Repository\nncase\src\NnCase.Converter\Converters\CaffeToGraphConverter.cs:line 47
at System.Linq.Enumerable.SelectIListIterator2.MoveNext() at System.Linq.Enumerable.WhereEnumerableIterator1.ToList()
at System.Linq.Enumerable.ToList[TSource](IEnumerable`1 source)
at NnCase.Converter.Converters.CaffeToGraphConverter.Convert() in D:\Work\Repository\nncase\src\NnCase.Converter\Converters\CaffeToGraphConverter.cs:line 29
at NnCase.Cli.Program.Main(String[] args) in D:\Work\Repository\nncase\src\NnCase.Cli\Program.cs:line 97
at NnCase.Cli.Program.

不支持数据读取层 怎么解决呢？ @sunnycase

With previous nncase, I had a workflow to use it to convert model to k210code and compile kendryte-standalone-sdk. Now nncase only ouputs k210modeo. What is the workflow to build a bin file?

Create a graph with input node 256x256x3, an output node 128x128x32, 3x3 kernel, and stride=2 Example tflite attached - detect.tflite.zip

nncase fails with Fatal: Allocator has ran out of memory

Result: The output node has shape: m_data = {1, 32, 256, 256}
Expected: Output node shape: 1x32x128x128

GDB log:

(gdb) b /home/tj/nncase/src/scheduler/freelist.cpp:69
Breakpoint 1 at 0x5555556738ed: file /home/tj/nncase/src/scheduler/freelist.cpp, line 69.
(gdb) run compile test/detect.tflite test/out.kmodel -i tflite --dataset test/test_000_0000000.png
Starting program: /home/tj/nncase/out/bin/ncc compile test/detect.tflite test/out.kmodel -i tflite --dataset test/test_000_0000000.png
[Thread debugging using libthread_db enabled]
Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1".
[New Thread 0x7ffff5714700 (LWP 17532)]
[New Thread 0x7fffecf13700 (LWP 17533)]
[New Thread 0x7ffff4f13700 (LWP 17534)]
[New Thread 0x7fffeffff700 (LWP 17535)]
[New Thread 0x7fffef7fe700 (LWP 17536)]
[New Thread 0x7fffeeffd700 (LWP 17537)]
[New Thread 0x7fffee7fc700 (LWP 17538)]

Thread 1 "ncc" hit Breakpoint 1, nncase::scheduler::freelist::reserve (this=0x5555562a0090, size=2097152) at /home/tj/nncase/src/scheduler/freelist.cpp:69
69 throw std::runtime_error("Allocator has ran out of memory");
(gdb) backtrace
#0 nncase::scheduler::freelist::reserve (this=0x5555562a0090, size=2097152) at /home/tj/nncase/src/scheduler/freelist.cpp:69
1 0x0000555555673760 in nncase::scheduler::freelist::allocate (this=0x5555562a0090, size=2097152) at /home/tj/nncase/src/scheduler/freelist.cpp:42
2 0x0000555555672973 in nncase::scheduler::memory_allocator::allocate (this=0x5555562a0080, size=2097152) at /home/tj/nncase/src/scheduler/memory_allocator.cpp:68
3 0x000055555566af2b in nncase::scheduler::allocation_context::allocate_default (this=0x7fffffffd710, conn=...) at /home/tj/nncase/src/scheduler/scheduler.cpp:50
4 0x000055555566b2d6 in nncase::scheduler::<lambda(nncase::ir::node&)>::operator()(nncase::ir::node &) const (closure=0x7fffffffd600, node=...) at /home/tj/nncase/src/scheduler/scheduler.cpp:82
5 0x000055555566bd03 in nncase::ir::relay_ir_visitor<nncase::ir::dfs_ir_visitor, nncase::scheduler::schedule(tcb::spannncase::ir::output_node*, nncase::scheduler::allocation_context&, std::vectornncase::ir::node*&)::<lambda(nncase::ir::node&)> >::visit(nncase::ir::node &) (this=0x7fffffffd5c0, node=...) at /home/tj/nncase/src/ir/include/ir/visitor.h:67
6 0x0000555555d4bf6d in nncase::ir::dfs_ir_visitor::visit_strategry (this=0x7fffffffd5c0, node=...) at /home/tj/nncase/src/ir/visitor.cpp:66
7 0x0000555555d4bf2e in nncase::ir::dfs_ir_visitor::visit_strategry (this=0x7fffffffd5c0, node=...) at /home/tj/nncase/src/ir/visitor.cpp:61
8 0x0000555555d4bf2e in nncase::ir::dfs_ir_visitor::visit_strategry (this=0x7fffffffd5c0, node=...) at /home/tj/nncase/src/ir/visitor.cpp:61
9 0x0000555555d4bf2e in nncase::ir::dfs_ir_visitor::visit_strategry (this=0x7fffffffd5c0, node=...) at /home/tj/nncase/src/ir/visitor.cpp:61
10 0x0000555555d4bd37 in nncase::ir::ir_visitor::visit (this=0x7fffffffd5c0, outputs=...) at /home/tj/nncase/src/ir/visitor.cpp:31
11 0x000055555566b895 in nncase::scheduler::schedule (outputs=..., context=..., compute_sequence=std::vector of length 1, capacity 1 = {...}) at /home/tj/nncase/src/scheduler/scheduler.cpp:123
12 0x00005555555d5905 in (anonymous namespace)::gencode (target=..., graph=..., options=...) at /home/tj/nncase/src/cli/compile.cpp:212
13 0x00005555555d7234 in compile (options=...) at /home/tj/nncase/src/cli/compile.cpp:265
14 0x00005555555a1222 in main (argc=8, argv=0x7fffffffe408) at /home/tj/nncase/src/cli/cli.cpp:38
(gdb) f 3
3 0x000055555566af2b in nncase::scheduler::allocation_context::allocate_default (this=0x7fffffffd710, conn=...) at /home/tj/nncase/src/scheduler/scheduler.cpp:50
50 auto &node = allocator->second->allocate(size);
(gdb) p conn
$1 = (nncase::ir::output_connector &) @0x55555629adc0: {nncase::ir::base_connector = {owner = @0x55555629ab10, name = "output", type_ = nncase::dt_uint8, shape_ = {static alignment = 8,
m_allocator = {<gnu_cxx::new_allocator> = {}, }, m_begin = 0x55555629ae10, m_end = 0x55555629ae30, m_capacity = 0x55555629ae30, m_data = {1, 32, 256,
256}}}, connections = std::vector of length 1, capacity 1 = {0x555556299ce0}, memory_type = nncase::mem_k210_kpu}
(gdb) c
Continuing.
Fatal: Allocator has ran out of memory
[Thread 0x7fffedffb700 (LWP 17596) exited]
[Thread 0x7fffeffff700 (LWP 17592) exited]
[Thread 0x7ffff7fd4740 (LWP 17586) exited]
[Thread 0x7fffee7fc700 (LWP 17595) exited]
[Thread 0x7fffeeffd700 (LWP 17594) exited]
[Thread 0x7fffef7fe700 (LWP 17593) exited]
[Thread 0x7ffff4f13700 (LWP 17591) exited]

Maybe related: #47 (comment)

would be helpful to remove the dimensions or axes that have a length of one

Hi,
I use mobilenet keras 0.5 and convert it to kmodel but I got error:

uasge: ./tflite2kmodel.sh xxx.tflite
Fatal: Invalid dataset.
System.ArgumentException: Invalid dataset.
at NnCase.Converter.Data.Dataset..ctor(String path, IReadOnlyCollection1 allowdExtensions, ReadOnlySpan1 dimensions, Int32 batchSize, PostprocessMethods postprocessMethod, Nullable1 mean, Nullable1 std) in D:\Work\Repository\nncase\src\NnCase.Converter\Data\Dataset.cs:line 67
at NnCase.Converter.Data.ImageDataset..ctor(String path, ReadOnlySpan1 dimensions, Int32 batchSize, PreprocessMethods preprocessMethods, PostprocessMethods postprocessMethod, Nullable1 mean, Nullable`1 std) in D:\Work\Repository\nncase\src\NnCase.Converter\Data\Dataset.cs:line 165
at NnCase.Cli.Program.Main(String[] args) in D:\Work\Repository\nncase\src\NnCase.Cli\Program.cs:line 225
at NnCase.Cli.Program.

Thanks

我看到项目描述中有对PReLU的支持，但今天在编译模型时无法编译 tflite 中的 ReLU 。我尝试将原模型中的 ReLU 转换为 PReLU 后在用 tf.lite 转换模型时还是会把PReLU转换为ReLU，应该怎么办。

Hi,
I convert tflite to kmodel with command:
./ncc/ncc -i tflite -o k210model --dataset images $1 ./$name

this is error:

uasge: ./tflite2kmodel.sh xxx.tflite
Fatal: Layer SUB is not supported
NnCase.Converter.Converters.LayerNotSupportedException: Layer SUB is not supported
at NnCase.Converter.Converters.TfLiteToGraphConverter.ConvertOperator(Operator op) in D:\Work\Repository\nncase\src\NnCase.Converter\Converters\TfLiteToGraphConverter.cs:line 63
at System.Linq.Enumerable.SelectEnumerableIterator2.ToList() at System.Linq.Enumerable.ToList[TSource](IEnumerable1 source)
at NnCase.Converter.Converters.TfLiteToGraphConverter.Convert() in D:\Work\Repository\nncase\src\NnCase.Converter\Converters\TfLiteToGraphConverter.cs:line 34
at NnCase.Cli.Program.Main(String[] args) in D:\Work\Repository\nncase\src\NnCase.Cli\Program.cs:line 110
at NnCase.Cli.Program.

thanks

i tried to do inference for mnist lenet using the night build:
nncase_night/ncc -i k210model -o inference --dataset ./images ./lenet2.kmodel ./output

I got this error:
Fatal: Layer Quantize is not supported
NnCase.Converter.Converters.LayerNotSupportedException: Layer Quantize is not supported
at NnCase.Converter.K210.Emulator.K210Emulator.Run(Tensor`1 batch, K210Conv2dLayerArgument inputArgument) in C:\projects\nncase\src\NnCase.Converter.K210\Emulator\K210Emulator.cs:line 106
at NnCase.Converter.K210.Emulator.K210Emulator.RunAsync(String datasetPath, String outputPath) in C:\projects\nncase\src\NnCase.Converter.K210\Emulator\K210Emulator.cs:line 59
at NnCase.Converter.K210.Emulator.K210Emulator.RunAsync(String datasetPath, String outputPath) in C:\projects\nncase\src\NnCase.Converter.K210\Emulator\K210Emulator.cs:line 55
at NnCase.Cli.Program.Main(String[] args) in C:\projects\nncase\src\NnCase.Cli\Program.cs:line 304
at NnCase.Cli.Program.

Convert TFLite model to K210 model.
Type the command: ncc -i tflite -o k210model ./detect.tflite ./detect.kmodel
An error: Fatal: Length of memory (162) must match product of dimensions (648).

@sunnycase 呼叫作者大大， 现在RELU也是不支持吗？

Hello guys,
we just try to convert our tflite model and got this error:
Fatal: Not supported tflite opcode: LOGISTIC
In the readme, LOGISTIC is one of the ops supported by the toolkit. I had tried on both pre-release v0.2.0 and release v0.1.5 rc5 but got same error,
Hope to receive feedback from you guys soon,
Thanks in advance

Hi,

I do not understand images\： quantization dataset images . What is it and how I can get it ???

Thanks

感谢您提供的工具。在readme里面，说是可以用ncc -i caffe *** 来转换caffe的模型。但是我从release下载的工具名字叫toco.exe，而不是ncc。另外使用toco.exe --input_format="caffe"报错，说不支持的格式。请问使用toco.exe如何将caffe模型转化成kmodel模型？其实虽然caffe现在没落了，但是很多企业由于历史原因，计算机视觉这块还在用caffe，所以对caffe模型的支持还是比较重要的。谢谢，盼复。

Could you please supply a CMakeLists.txt for compiling your yolo_v2 demo?

when I compiling the project, the error occurred as follows:
CMake Error: The source directory "X:/kendryte/nncase-master/examples/20classes_yolo/k210/kpu_20classes_example" does not appear to contain CMakeLists.txt. Specify --help for usage, or press the help button on the CMake GUI.

hu@hu-VirtualBox:~/ws$ nncase_night/ncc -i tflite -o k210model lenet01.tflite lenet01.kmodel
Fatal: Layer tflite.Operator is not supported: Only scalar multiply is supported
NnCase.Converter.Converters.LayerNotSupportedException: Layer tflite.Operator is not supported: Only scalar multiply is supported
at NnCase.Converter.Converters.TfLiteToGraphConverter.ConvertMul(Operator op) in C:\projects\nncase\src\NnCase.Converter\Converters\TfLiteToGraphConverter.cs:line 243
at NnCase.Converter.Converters.TfLiteToGraphConverter.ConvertOperator(Operator op) in C:\projects\nncase\src\NnCase.Converter\Converters\TfLiteToGraphConverter.cs:line 79
at System.Linq.Enumerable.SelectEnumerableIterator2.ToList() at System.Linq.Enumerable.ToList[TSource](IEnumerable1 source)
at NnCase.Converter.Converters.TfLiteToGraphConverter.Convert() in C:\projects\nncase\src\NnCase.Converter\Converters\TfLiteToGraphConverter.cs:line 34
at NnCase.Cli.Program.Main(String[] args) in C:\projects\nncase\src\NnCase.Cli\Program.cs:line 113
at NnCase.Cli.Program.

ncc.exe -i k210model -o inference --dataset ./images ./lenet2.kmodel ./output

Fatal: input-format
System.ArgumentException: input-format at NnCase.Cli.Program.Main(String[] args) in D:\Work\Repository\nncase\src\NnCase.Cli\Program.cs:line 117
at NnCase.Cli.Program.

Hi, when I converted TFLite model(mobilenet_ssd) to K210 model, I got an error:Fatal: Index was outside the bounds of the array. What may lead to this error? thanks!

Hi, Is there a simulator to run kmodel without physical HW?

Thanks

I'm doing a tutorial on making this kmodel
https://bbs.sipeed.com/t/topic/682

When proceeding in this way, the error mentioned in the title occurred

# ./tflite2kmodel.sh /tf/work/Maix-Keras-workspace/mbnet/mbnet75.tflite
uasge: ./tflite2kmodel.sh xxx.tflite
Fatal: Invalid dataset.
System.ArgumentException: Invalid dataset.
   at NnCase.Converter.Data.Dataset..ctor(String path, IReadOnlyCollection`1 allowdExtensions, ReadOnlySpan`1 dimensions, Int32 batchSize, PostprocessMethods postprocessMethod, Nullable`1 mean, Nullable`1 std) in D:\Work\Repository\nncase\src\NnCase.Converter\Data\Dataset.cs:line 67
   at NnCase.Converter.Data.ImageDataset..ctor(String path, ReadOnlySpan`1 dimensions, Int32 batchSize, PreprocessMethods preprocessMethods, PostprocessMethods postprocessMethod, Nullable`1 mean, Nullable`1 std) in D:\Work\Repository\nncase\src\NnCase.Converter\Data\Dataset.cs:line 212
   at NnCase.Cli.Program.Main(String[] args) in D:\Work\Repository\nncase\src\NnCase.Cli\Program.cs:line 249
   at NnCase.Cli.Program.<Main>(String[] args)

I would be grateful if you could give me advice on what to check.

Below is the model I created

https://drive.google.com/file/d/1W2u1D2zJAmnrNlqFqWP6E6l9QxpvQJyY/view?usp=sharing

2019-05-03 12:05:56.214925: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
Fatal: Layer DepthwiseConv2d is not supported
NnCase.Converter.Converters.LayerNotSupportedException: Layer DepthwiseConv2d is not supported
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.g__ConvertLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 51
at NnCase.Converter.K210.Converters.Stages.Convert.Converter.Convert(Graph graph, QuantizationContext quantizationContext, Int32 weightsBits) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\Stages\Convert\Converter.cs:line 60
at NnCase.Converter.K210.Converters.GraphToK210Converter.ConvertAsync(Dataset dataset, GraphPlanContext planContext, String outputDir, String prefix, Boolean channelwiseOutput) in C:\projects\nncase\src\NnCase.Converter.K210\Converters\GraphToK210Converter.cs:line 38
at NnCase.Cli.Program.Main(String[] args) in C:\projects\nncase\src\NnCase.Cli\Program.cs:line 246
at NnCase.Cli.Program.

Hello, When trying to convert my tflite model, I'm running into the following error:

Fatal: Dimensions must be equal. System.InvalidOperationException: Dimensions must be equal.    at NnCase.Converter.Model.InputConnector.SetConnection(OutputConnector from) in D:\Work\Repository\nncase\src\NnCase.Converter\Model\InputConnector.cs:line 32    at NnCase.Converter.K210.Transforms.K210SeparableConv2dTransform.Process(TransformContext context) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Transforms\K210SeparableConv2dTransform.cs:line 89    at NnCase.Converter.Transforms.Transform.Process(Graph graph, IReadOnlyList`1 transforms) in D:\Work\Repository\nncase\src\NnCase.Converter\Transforms\Transform.cs:line 77    at NnCase.Cli.Program.Main(String[] args) in D:\Work\Repository\nncase\src\NnCase.Cli\Program.cs:line 235    at NnCase.Cli.Program.<Main>(String[] args) ``` 

What could be the source of the problem? 
Thank you for your help.

using this https://github.com/kendryte/nncase/releases/tag/v0.1.0-rc5, drag and dropping it in ./ncc
and calling ncc with ./ncc/ncc
I'm getting this error
Could not resolve CoreCLR path. For more details, enable tracing by setting COREHOST_TRACE environment variable to 1

您好，我改变输入，但模型的输出始终没有变化, 是否我的模型转换为KMODEL后有问题？

Fatal: Exception has been thrown by the target of an invocation.

这个issue #11 中提到的nightly-build似乎没有可直接下载的链接。求官方放出预编译的macOS版。

I tried to compile the model from tflite to kmodel on Google Colaboratory..

./ncc/ncc compile /content/model.tflite /content/model.kmodel -i tflite -o kmodel --dataset /content/test/

But ncc output the error.

Fatal: Shapes must be same, but got [1x1x1x1] and [1x1x1x7]

How can I compile successfully?

nncase is failing when using a graph with intermediate tensors in: 32x128x128 out: 64x128x128.

Sample tflite graph

./ncc compile detect.tflite detect.kmodel --dataset images/ -i tflite --dump-ir
Fatal: Allocator has ran out of memory

32x128x128 + 64x128x128 = 1048576 + 524288 = 1,572,864

I'd expect this to work since it's less than 2,097,152 bytes.

ir_optimize_2:

ir_import:

Hi,

how can I build this code ?

Thanks

While converting tflite to kmodel, I am encountering out of memory. I still have 2.6GB ram left. Any comment?

2019-08-03 00:41:10.769773: I tensorflow/core/platform/cpu_feature_guard.cc:141] Your CPU supports instructions that this TensorFlow binary was not compiled to use: SSE4.1 SSE4.2 AVX AVX2 FMA
0: InputLayer -> 1x3x224x224
1: K210Conv2d 1x3x224x224 -> 1x24x112x112
2: K210Conv2d 1x24x112x112 -> 1x24x112x112
3: K210Conv2d 1x24x112x112 -> 1x16x112x112
4: K210Conv2d 1x16x112x112 -> 1x96x112x112
Fatal: KPU ran out of memory.
System.InvalidOperationException: KPU ran out of memory.
   at NnCase.Converter.K210.Converters.Stages.Inference.KPUMemoryAllocator.Allocate(UInt32 size) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\KPUMemoryAllocator.cs:line 28
   at NnCase.Converter.K210.Converters.Stages.Inference.InferenceContext.GetOrAllocateKPUMemory(OutputConnector output) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferenceContext.cs:line 40
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.AllocateInputMemoryDefault(Layer layer, OutputConnector input, InferenceContext context) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 123
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 50
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.<Infer>g__InferLayer|0_6(Layer layer, <>c__DisplayClass0_0& ) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 41
   at NnCase.Converter.K210.Converters.Stages.Inference.InferExecutor.Infer(Graph graph, ConvertContext convertContext) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\Stages\Inference\InferExecutor.cs:line 106
   at NnCase.Converter.K210.Converters.GraphToK210Converter.ConvertAsync(Dataset dataset, GraphPlanContext planContext, String outputDir, String prefix, Boolean channelwiseOutput) in D:\Work\Repository\nncase\src\NnCase.Converter.K210\Converters\GraphToK210Converter.cs:line 41
   at NnCase.Cli.Program.Main(String[] args) in D:\Work\Repository\nncase\src\NnCase.Cli\Program.cs:line 271
   at NnCase.Cli.Program.<Main>(String[] args)```

nice work，thanks！
Why do we need to set up dataset for k210 model? Is it to set the size of network entrance?
What should I do when I take pictures directly from the camera as input ？