Ethon Lin
Ethon Lin
About
- Username
- Ethon Lin
- Joined
- Visits
- 1,233
- Last Active
- Roles
- Member, ModeratorsKL520, ModeratorsKL720, ModeratorsAIModel
Comments
-
Hi Jerome, The latest version of toolchain is v0.12.1, you can convert and compile the model yolov3 on it. Please refer to the instruction manual. (http://doc.kneron.com/docs/#toolchain/manual_520/ ) The toolchain only supports opset = 9 now, optset…
-
https://www.kneron.com/forum/discussion/comment/144#Comment_144 Hi WenTzu, What kind of image format are you using? General format like jpg and png are usable. Or could you post your json setting and onnx to clarify the issue?
-
Hi Jerome, There are some version condition in toolchain. Take pytorch as an example, please check the version are "pytorch = 1.4" and "onnx = 1.4" . When export model to onnx from pytorch, make sure the opset version equals 9. Y…
-
Hi Alfred, 看過你的模型後，有幾個KL520不支援的地方需要請你修改模型，請參考下面步驟 https://www.kneron.com/forum/uploads/051/1L1ZIOX1K1LU.png 模型使用的opset版本為v8(如上圖右側紅圈處所示)，但KL520目前僅支援v9，還請在轉成onnx時選擇v9 KL520並不直接支…
-
是的，需先將模型轉成KL520專用的*.nef (舊版的範例中為*.bin) 後才可在KL520上運行，轉換的方式可以參照文件中心http://doc.kneron.com/docs/#toolchain/manual_520/ 的說明 成功將模型轉換成.onnx後，請繼續參照 3.2 FpAnalyser, Compiler and IpEvaluator 的指示依…
-
Yes, that's what I'm getting at. If you want to get the same model result, please calculate the layer softmax with the inf_res from kdp_dme_retrieve_res(), the prediction scale should be normalized after implement softmax.
-
The score range of final prediction should be the same as your original model. And there is usually a layer "softmax" in the end of MobileNetV1, but KL520 doesn't support "softmax". So I believe that "softmax" was be r…
-
The hardware structure is using RGBA8888 for 4 channels, but if you use the input format RGB565, you will get better accuracy with the model trained by image RGB565. And for better accuracy, there are two ways you can try. Modify the parameter NPU_…
-
Here is the format of RGBA8888, please check whether your data format is correct. https://www.kneron.com/forum/uploads/518/C2O5P2QAWFAH.png And may I ask your setting "img_preprocess_method" of input_params.json in toolchain?
-
Hi Paolo, The issue was caused by the scpu fw error with the binary fw_scpu.bin in folder age_gender. Please copy the "fw_scpu.bin" in /host_lib_v0.9.0-early/app_binaries/KL520/tiny_yolo_v3 into /host_lib_v0.9.0-early/app_binaries/KL520/o…
-
Hi Paolo, Yes, now the App. age_gender only returns result with the face which is the highest score. If you just want to detect where the faces are, please refer to the example ssd_fd and you can get result with all the faces.
-
If you worry about the data loss by format transferring, you can try to choose the format NPU_FORMAT_RGBA8888. And to reduce quantization loss during model was transferring to fixed-point structure, make sure there are over 100 images in the folder…
-
Due to the NPU structure, the length of width should be round up to 16, and there should be 6 floating(4bytes) head numbers in the forefront of the output array. In your case, the total length should be 6 * 4 + 3 * round_up16(1) = 72 …
-
Thanks for your suggestion. We will make modification to it.
-
Due to the "inf_size" = 0, kdp_dme_retrieve_res could not get correct length to receive result data. Refer to the return value of api.kdp_dme_inference below: ( You can find it in __init__.py of kdp_host_api) return ret, inf_size_p…
-
For business policy, we no longer provide full application code of "age_gender" for free. But you can still find the model binaries in host_lib example (\host_lib_v0.9.0-early\app_binaries\KL520\age_gender). (Download host_lib_v0.9.0-early…
-
Yes, if the inference process is done, there should be some results whether it's right or wrong. Could you mind to provide your model binaries and python example code to check the issue?
-
Kneron provided firmware sdk for all KL520 device. Please download SDK from Development Center (https://www.kneron.com/tw/support/developers/) or get sdk v1.3 from following link. https://www.kneron.com/tw/support/developers/?folder=KL520%20SDK/&…
-
Due to some reasons, here are few operations that we are recommended you to try. Update the fw binaries in \host_lib__v0.8\app_binaries\tiny_yolo_v3: The fw and host_lib should be match to ensure the function works, and we're recommended to use th…
-
The mode = 0 for dme_inference means serial mode, in serial mode, just use the api "kdp_dme_retrieve_res" with the pointer "inf_res" to get the inference result if the return value of inference "_ret" is 0. e.g. api.kdp…
-
https://www.kneron.com/forum/discussion/comment/18#Comment_18 The radix setting in input_params.json is just for simulator which is a software simulating function of Toolchain. To set the radix = 7 in actual NPU inference, you should add the action…
-
https://www.kneron.com/forum/discussion/comment/15#Comment_15 Sorry for the confusion. With the radix = 0 and there is a prefix layer BatchNormalization in your model, there are some correspond conditions you should modify before model inference wit…
-
Here are constraints of dongle KL520 a. Flash size: 32MB, model should be smaller than 120MB so could be compressed to under 30MB in fixed-point structure. b. Supported operators: please refer to the chapter2.3 in the document center http://doc.kner…
-
https://www.kneron.com/forum/discussion/comment/12#Comment_12 That's because the NPU is based on a int8 structure, if the input data is over the boundary of int8, it will make some error inside the NPU. For the tensorflow preprocess (-1~1), to chang…
-
The warning message < [common][warning][operation_node_gen.cc:640][AddReformatPrefix] Input format in configrgba8 is RGBA but does not match format of first layerseq8, add identical prefix conv node to handle > was shown due to the operator &…
-
Here is the document of toolchain, please install docker and you can pull the toolchain image. http://doc.kneron.com/docs/#manual_520/