Maria Chen
Maria Chen
About
- Username
- Maria Chen
- Joined
- Visits
- 1,092
- Last Active
- Roles
- Member, ModeratorsKL520, ModeratorsKL720, ModeratorsAIModel
Comments
-
Hi Richard, 在toolchain上透過IP evaluator後可以得到onnx的報告(如圖)，由報告結果可看到NPU time: 25ms, 代表NPU的部份推論時間為25ms, 即使加上前處理5ms，理論上也可以達到 30 FPS ，10.75 FPS的速度推測應該是指整個系統串接後的結果。速度變慢的可能原因有很多(…
-
Hi Pin, -boot_spl.bin在KL520 SDK裡面，可以直接使用 (在KL520_SDK的utils/bin_gen/flash_bin裡) -fw_ncpu.bin和fw_scpu.bin在Kneron PLUS的firmware資料夾裡面，如果要改firmware，需要下載Keil並購買Keil的license，就可以用Keil改KL520 SDK的firmware後build新的fw…
-
您好, 那個error還有另外一個可能性: 因為您的「下載」資料夾是用中文寫的，所以有可能program找不到檔案。可以請您試著把Kneron PLUS的路徑放在沒有中文的地方，再試一次嗎? 如果還是不行，請確認Jetson Xavier NX的ubuntu platform有沒有符合x86_64 (不支援ARM。這時…
-
您好， 用tensorflow訓練出來的模型，要轉換成.onnx (有指定版本)，再轉換成.nef檔，就可以用加速棒去跑。我們有提供轉成.onnx檔與.nef檔的script: ONNX Converter - Document Center (kneron.com)。 轉換模型時，請注意KL520與KL720環境支持的operators: Toolchain Ma…
-
您好， 耐能的產品會幫助AI去做運算與提供結果，有些app可以輸出圖片或顯示相機的螢幕，有些app只會印出結果(沒有UI)。執行這些app會需要input(圖片)跟模型跟firmware (可以下載Kneron PLUS看一下範例，這裡有說明: Index - Document Center (kneron.com))。您的app可…
-
Hi Su, 在執行gereric inference時不用加sudo，因為在python環境中用sudo的話會跑到另外一個環境，可能就不是您安裝的環境了。 若是沒有用sudo會遇到USB權限問題，請參考文件說明設定使用權限: Install Dependency - Document Center (kneron.com) 還有，您可以確認是…
-
Hi Su, 可以試試看這個連結嗎? Install Dependency - Document Center (kneron.com) 我剛剛用您的連結，發現連不上，不過拿掉https裡面的s之後，就可以連上了。
-
Hi Anoop, Where is your model from, and how did you train your model? If you didn't follow Kneron mmlab for model training and converted a self-trained model to .nef, you would need to inference using Kneron PLUS 2.0.1 and write your own generic inf…
-
Hi Anoop, Thank you for providing the files. Here are a few things you could check: -When you generate the .nef file, please check if your normalization settings match the settings used in device inferencing. For example, if your configuration for c…
-
Hi Hunter, The FPS also depends on your image size and image format, but we support 2 MIPI lanes, and the speed is 2.5Gbps per lane. The hardware supports resolution of maximum 8192 pixels horizontally.
-
Hi Dennis, I'm glad the colors look okay now!! Thanks for letting us know how AWB and AEC would affect the colors, I didn't know that! And no problem!
-
Hi Anoop, Could you provide us the inference program and the image used, along with your NEF file? Thanks for your help. By the way, if you set to_rgb = True in your configuration file when you trained your model, please make sure you aren't convert…
-
Hi Dennis, No problem!! I also use something similar to you to read USB: https://www.kneron.com/forum/uploads/433/48F9FBV70NAH.png Then, I use msg_in to read 640*480*2. I keep the parameters the same on the firmware side: usb_com_write(p_img_buf, IM…
-
您好， 您的理解大概都是正確的。.onnx檔案之後需要轉換成.nef檔，就可以用加速棒加上Kneron的Python API運行喔 (Kneron PLUS裡面有Python範例可以參考) 我們的產品沒有特別說支援哪些模型，不過我們有NPU支援的operator list，若模型中使用的都是NPU有支援的，那就可…
-
Hi Anoop, We ran your onnx model and got the same error as you when we didn't go through ktc.onnx_optimizer.onnx2onnx_flow. We optimized your onnx file and successfully generated .nef files for both KL520 and KL720. The code below worked for us: htt…
-
Hi Dennis, I think I see your point now. The image on the PC and the display on the DevKit looks like they have different colors. When I tried it, the white parts also looked a little yellow, and the red parts' saturations looked a little different …
-
Hi Richard, 對，您可以嘗試把模型輸入的解析度改小。Feature map的大小會因為模型輸入的解析度縮小而變小喔。
-
Hi Dennis, If you were using usb.back.libusb1.get_backend(...) and usb.core.find(...) to connect to the KL520 USB, how do you get the image data transmitted from your USB? I'm getting the data by using: https://www.kneron.com/forum/uploads/128/CZMES…
-
Hi Richard, 謝謝您提供的檔案。我們的同事確認了，這個模型的確是超過了硬體的限制。KL720的NPU有個DRAM address，有16MB的限制。 您可以嘗試把feature map的size改小。
-
Hi Dennis, Thank you for providing your code. Since my code is in C, it's taking some time for me to reproduce the issue. I'm not sure if the AWB, AEC, and gamma effects would matter. Have you tried turning them on to see what the image looks like? …
-
您好， 我們有提供mmdetection的流程給您參考: kneron-mmdetection/yolox_step_by_step.md at main · kneron/kneron-mmdetection · GitHub 在Step 1-2用tools/train.py訓練模型後生成的latest.pth模型檔，可以從Step 2繼續下去測試和轉換，之後再到Step5用toolchain來…
-
Hi Richard, 這個訊息很有可能是來自於硬體的限制。 如果您可以提供.nef檔，.onnx檔，還有轉檔的script，我們可以幫忙確認。
-
Hi Anoop, It seems like this model hasn't gone through ONNX to ONNX. Please refer to Toolchain Manual - Document Center (kneron.com), sections 3.1.2 Pytorch to ONNX and 3.1.5 ONNX Optimization. After using the Python API to go through these 2 steps,…
-
Hi, 推論的圖片格式與模型轉換的圖片格式沒有關係，所以img_channel設定為RGB就可以了喔。 推論時，請將image format設定成KP_IMAGE_FORMAT_YUYV。 https://www.kneron.com/forum/uploads/306/WRPQ7FYUF9PU.png
-
Hi Anoop, Could you run "eval_result = km.evaluate()"? If so, what does the result say? There are some possible reasons, such as the input shape is incorrect, or there is an unsupported operator, but we would need more information to find …
-
Hi Dennis, Is the SDK version you're using KL520 SDK v1.7.0? Also, are you using OpenCV on your PC? If so, OpenCV's default color format is BGR, and the color format KL520 SDK uses is RGB565, so you might want to convert your image to BGR before dis…
-
Hi Su, KL720DemoGenericInferenceYoloX_BypassHwPreProc.py裡面的postprocessing是給YoloX用的，所以可能會有問題。 要用任何模型，可以去耐能的開發中心下載Kneron PLUS (開發中心 | Kneron - 人工智慧無所不在)，有一些C與Python範例可以跑，像是KL720DemoGenericIm…
-
您好， 太好了! 很高興訓練模型指令可以正常運作。 這個FileNotFoundError，是因為dataset的實際路徑是data/imagenet/val/O/O_xxx.jpg (我是在train的資料夾找到O_1004.jpg的)，圖片外面有O跟R的資料夾，所以這個路徑是錯的。 建議您可以把O與R的圖片，全都放在Val資料…
-
您好， 我們的產品沒有特別說支援哪些模型，不過我們有NPU支援的operator list，若模型中使用的都是NPU有支援的op，那就可以在KL520上面執行。 https://www.kneron.com/forum/uploads/319/94E0OU134SH3.png Operator list: Toolchain Manual - Document Center (kneron…
-
您好! 我現在下載您的dataset確認了，可以正常訓練 (但是Validate時會有FileNotFoundError)。 所以不是dataset的問題，不好意思! 在regnetx-400mf_8xb128_in1k.py裡，我只有改samples_per_gpu與workers_per_gpu，也有在kneron-mmclassification裡新增data與imagenet資…