Ksnn changelog required

Hi @Frank
I saw some recent commits and a new minor version release for ksnn. Could you please provide a changelog for the same? Thanks

@johndoe

  1. Add RTSP Web Camera Demo for Yolo series
  2. Fix some bugs
  3. Adapted to NPU 6.4.6.2
1 Like

Could you please mention the bugs that were fixed?

Also, there’s a commit with message input api: add more input data format: i8, i16, f16. Could you please elaborate what additional functionality this provides?

@johndoe When the input of your model itself is not u8, there will be a segmentation fault after conversion. I fixed this problem and added different inputs

Got it. I think this was the cause of a related issue`

Are there any additional parameters that we need to pass (input data type) during conversion?

@johndoe This is determined by your model itself and cannot be modified

1 Like

I’m facing this error after trying to run inference with a model quantised with dynamic-fixedpoint int16

#productname=VIPNano-QI, pid=0x88
Create Neural Network: 664ms or 664625us
Traceback (most recent call last):
  File "resnet18.py", line 95, in <module>
    outputs = resnet.nn_inference(img_list, input_tensor=1, output_tensor=1, platform = 'ONNX', reorder='2 1 0', output_format=output_format.OUT_FORMAT_FLOAT32)
  File "/tmp/goldenmask33u69bkv/ksnn/api.py", line 234, in nn_inference
  File "/tmp/goldenmask33u69bkv/ksnn/api.py", line 149, in nn_set_inputs
  File "/tmp/goldenmask33u69bkv/ksnn/kplatform/vim3/vim3_nn_api.py", line 191, in neural_network_set_input_i16
  File "/usr/lib/python3.8/ctypes/__init__.py", line 386, in __getattr__
    func = self.__getitem__(name)
  File "/usr/lib/python3.8/ctypes/__init__.py", line 391, in __getitem__
    func = self._FuncPtr((name_or_ordinal, self))
AttributeError: /home/khadas/test_01/outputs/onnx/2/inception-v1/libnn_inception-v1.so: undefined symbol: NPU_SetInput_I16

@johndoe Can you share me you model ? I will test it . Maybe some bug here

@Frank Sure. Here’s it converted model

The original model is taken from torchvision

The conversion script is as follows:

./convert \
--model-name resnet18-v1 \
--platform onnx \
--model onnx_models/resnet18-v1.onnx \
--mean-values '103.94,116.78,123.68,58.82' \
--quantized-dtype dynamic_fixed_point --qtype int16 \
--kboard VIM3 --print-level 1

@Frank Were you able to figure out the issue?

@johndoe You need to wait, there is no time to test this now, I will find time to test it

@johndoe Did you update you SDK to 0fb310abaa9b28d5f08a8cd9e747422f296db32e ?

Try those commands

$ git pull origin master
$ git submodule init
$ git submodule update
1 Like

I’m stuck with a recurring error link

Update: Working now after updating the SDK