Default Avatar

Eric Wu

Eric Wu

About

Username
Eric Wu
Joined
Visits
254
Last Active
Roles
Member, ModeratorsAIModel

Comments

  • 建議可以先確認訓練過程中的log數據皆正常合理, 如果出現不合理的log數據, 可以根據log數據中的發現, 對相應的訓練參數做嘗試與修改(加大訓練數據, 修正數據, 加大epoch ... etc).
    in YOLOv5 on KL720 toolchain Comment by Eric Wu November 2021
  • Hi Shih-Hsin Chen, 目前tool chain並沒有直接支援tensorflow.js的格式, 但我相信有許多第三方套件可以幫忙轉換你的tensorflow.js格式. 轉成onnx, tflite...etc 之後, 即可使用tool chain繼續.
    in KL520如何加速Tensorflow.js Comment by Eric Wu November 2021
  • @Jasmine 您提供的指令(compilerIpevaluator_720.sh) 與 圖片中的指令(fpAnalyserBatchCompile_720.py)貌似不太一樣, 您可能要在再確認一下. 我這邊嘗試使用 tftargetcut_v2_opt 做 fpAnalyserBatchCompile_720.py 與 compilerIpevaluator_720.sh 都可以正常執行. …
    in [Toolchain]若模型在轉換的過程中發生錯誤，該如何自我排除? Comment by Eric Wu November 2021
  • 關於 "該如何得知"輸出output node"為正確": 可以依照自己對這個模型的了解,以及接下來要做的事 (inference時, 沒轉出來的部分會需要以其他方式實現),  轉出來的model如果沒有你要的operator或與想像的不同(e.g. model size太小), 就是設定錯誤了…
    in [Toolchain]若模型在轉換的過程中發生錯誤，該如何自我排除? Comment by Eric Wu October 2021
  • @Jasmine 輸出output node時必須要設定正確, 除此之外, 這邊有一點tricky, 由於tensorflow輸出的model要做export時有時候會把常見的operator拆成奇怪的operator, 建議先把模型輸出的tflite後在做轉換會比較容易(tf官方有提供api做轉換, 且會處裡這些奇怪的opera…
    in [Toolchain]若模型在轉換的過程中發生錯誤，該如何自我排除? Comment by Eric Wu October 2021
  • Hi Jasmine, tensorflow 1.15.0應該可以使用 要找到node name, 用tensorboard可以, 或是直接印出來也可以:  https://stackoverflow.com/questions/36883949/in-tensorflow-get-the-names-of-all-the-tensors-in-a-graph
    in [Toolchain]若模型在轉換的過程中發生錯誤，該如何自我排除? Comment by Eric Wu October 2021
  • Hi Fred, 文章(https://www.kneron.com/forum/discussion/53/legacy-v0-12-example-keras-kl520-how-to-convert-and-compile-tiny-yolo-v3-from-github-project#latest) 是示範一個例子如何將github上的開源軟體中的pretrain model轉換成nef. 你在訓練時發生錯誤有可…
    in 訓練自己的模型時跳出KeyError: 'val_loss' Comment by Eric Wu October 2021
  • Hi Fred, 模型認不到物件可能原因有很多, bad training, input data, preprocess code,  postprocess code, quantization ... etc 建議先確認你的原始h5模型可以正確認出物件. 原始repository( keras-yolo3 )中應有對應的資源. 再驗證與onnx轉換正確可以正確認出物件.…
    in 使用自行編譯出的.nef進行KL520推論失敗 Comment by Eric Wu October 2021
  • Hi 夏英杰, 我猜可能是環境中的onnx版本不正確. 請確認環境中onnx的版本, 目前主要使用onnx==1.6. https://github.com/kneron/ONNX_Convertor#introduction 新版本的onnx已經移除optimizer模組: https://github.com/onnx/onnx/releases/tag/v1.9.0
    in 該如何使用其他的 onnx 檔? Comment by Eric Wu August 2021
  • Hi mint, your onnx need some modification, https://www.kneron.com/forum/uploads/158/CXF0CTWTJ9A4.png Our kl520 could not support Transpose node, you need to remove it. Our toolchain need input shape and all feature map shape information, "?&q…
    in Issue about converting onnx file to the nef file Comment by Eric Wu August 2021
  • Hi 楊堯順, 方便提供下列項目: failed onnx model 可以重置toolchain問題的python script toolchain version 以利我們定位錯誤原因嗎? thanks.
    in keras to onnx 問題 Comment by Eric Wu July 2021
  • https://www.kneron.com/forum/discussion/comment/301#Comment_301 Hi Raef Youssef, Yes, we train yolov5s and yolov5s_noupsample with focus layer(slice version). Only do convolution layer substitute when doing exporting to onnx for KL520.
    in Porting YOLOv5 Comment by Eric Wu June 2021
  • Hi Raef Youssef, the "model_training" document is for the script in "/workspace/ai_training" (folder path in toolchain:v0.14.2). And, in order to port focus layer to kl520, we use a "conv" instead of those "slice…
    in Porting YOLOv5 Comment by Eric Wu June 2021
  • Hi ChiYu, seems that your exported onnx is invalid, you can check your onnx with official onnx api as following: import onnx import onnx.shape_inference m = onnx.load('/data1/retina.onnx') onnx.checker.check_model(m) onnx.shape_inference.infer_s…
    in Inferred shape and existing shape differ in rank: (0) vs (3) (pytorch_exported_onnx2optimized_onnx) Comment by Eric Wu June 2021
  • https://www.kneron.com/forum/discussion/comment/284#Comment_284 Hi Raef Youssef, Yes, you can use relu6, but you can't multiply it by variable x because kl520 can't support "MUL" operator. for removing the preprocessing layer, currently …
    in Porting YOLOv5 Comment by Eric Wu May 2021
  • https://www.kneron.com/forum/discussion/comment/282#Comment_282 Hey Raef Youssef, Yes, KL520 support ReLU6 using Clip. Some operators in different deep learning framework have different name even though they have same behavior. And, onnx doesn't h…
    in Porting YOLOv5 Comment by Eric Wu May 2021
  • Hi @Tim Gilmanov , What toolchain version you use? the converted onnx you provided seems not match to our latest toolchain (v0.14): yours: https://www.kneron.com/forum/uploads/685/PTKZEYV1K0IS.png expected: https://www.kneron.com/forum/uploads/50…
    in Error when attempting model editing (removing the softmax) Comment by Eric Wu May 2021
  • Hello 林韋銘: to learn the python api usage, you can check the python api document: http://doc.kn eron.com/docs/#toolchain/python_api/ chapter 3. Inferencer and the real world case(yolo) tutorial: http://doc.kneron.com/docs/#toolchain/yolo_e…
    in AttributeError when converting onnx model using pytorch2onnx.py Comment by Eric Wu May 2021
  • https://www.kneron.com/forum/discussion/comment/267#Comment_267 Hi Ellen, 看起來是在post_process (yolo_eval) 的部分出問題, 請問你有試過在keras可以成功inference嗎? 你更改了模型參數, 可能跟tutorial的設置的參數不符, 需要根據你的需求作修改. 建議先確認…
    in Keras to ONNX problem Comment by Eric Wu May 2021
  • Hi Raef Youssef, there are several operators are not supported by kl520 in your yolo-v5, you can check the document (chapter 2.3 supported operators): http://doc.kneron.com/docs/#toolchain/manual/ if you want to port this yolo-v5 to kl520, you …
    in Porting YOLOv5 Comment by Eric Wu May 2021
  • Hi Ellen, 你的model含有Lambda layer, 我們並不支援這個type的轉換, 必須想辦法拿掉這個layer. 修改source code是一個方法, 可以參考tutorial中step 5 (remove redundant layer): https://www.kneron.com/forum/discussion/53/legacy-v0-12-example-keras-kl520-how…
    in Keras to ONNX problem Comment by Eric Wu May 2021
  • https://www.kneron.com/forum/discussion/85/tflite-model-migration Hi Raef Youssef, It's error comes from tflite intepreter while loading model, I guess there is a tensorflow version mismatch issue while convert tensorflow model to tflite model, Co…
    in Tflite Model Migration Comment by Eric Wu May 2021
  • https://www.kneron.com/forum/discussion/comment/232#Comment_232 Hi Raef Youssef, this is the issue comes from the changing of opset 9 to 11. when doing "torch.onnx.export" using opset_version=11, the torch give us the model with extra br…
    in Ultralytics Tiny YOLOv3 Comment by Eric Wu May 2021
  • Hi francisco soto, I saw there is "auto_pad" attribute in every convolution layer in the model you provided, https://www.kneron.com/forum/uploads/034/YSBBM29NUCD8.png I have some experience about a bug comes from onnx builtin function du…
    in Error when running FpAnalyser with tiny yolov4 Comment by Eric Wu April 2021
  • https://www.kneron.com/forum/discussion/comment/209#Comment_209 Hi Ellen, train_weights_final.h5跟yolov3-tiny.weights是不同的東西, train_weights_final.h5是keras格式的weight file. 使用netron查看的話看起來像這樣: https://www.kneron.com/forum/uploads/…
    in Keras to ONNX problem Comment by Eric Wu April 2021
  • Hi Phidias, Really thanks your post, we will fix this bug in next release.
    in [Bug] Wrong script path in /workspace/scripts/convert_model.py in latest toolchain docker image Comment by Eric Wu April 2021
  • https://www.kneron.com/forum/discussion/comment/200#Comment_200 Hi Ic.Wang, For the "/data1/images" I lost my calibration data in this case, so I do that again with the following dataset. It's still work. The dataset is picked from VOC 20…
    in [LEGACY][v0.12][Example][Keras][KL520] How to convert and compile tiny-yolo-v3 from github project Comment by Eric Wu April 2021
  • Just want to share a tricky way to convert, In this case, I tried to downgrade the torch version to match the required ir version in kneron/toolchain:520_v0.12.0 ( onnx: 1.4.1 ir_version: 4), but, I got model loading issue with low torch version …
    in Ultralytics Tiny YOLOv3 Comment by Eric Wu March 2021
  • https://www.kneron.com/forum/discussion/comment/150#Comment_150 Hi @Raef Youssef, I found there is a mistake in the export_tiny.py, we have to change the anchor and classes information file: https://www.kneron.com/forum/uploads/615/5UQN38T7UQO2.png …
    in [LEGACY][v0.12][Example][Keras][KL520] How to convert and compile tiny-yolo-v3 from github project Comment by Eric Wu March 2021
  • https://www.kneron.com/forum/discussion/comment/139#Comment_139 Hi @Raef Youssef , For the cpu node implementation, Sorry, I don't know the detail about this. Someone who know this part might reply this here. But for the upsampling layer, The up…
    in [LEGACY][v0.12][Example][Keras][KL520] How to convert and compile tiny-yolo-v3 from github project Comment by Eric Wu March 2021

© Kneron, Inc.