
sphuang
sphuang
About
- Username
- sphuang
- Joined
- Visits
- 55
- Last Active
- Roles
- Member
Comments
-
@Maria Chen Hi Maria 將上述四個參數修改為int8後此模型一樣是不能轉換 https://drive.google.com/file/d/1HS766EdjQQ8qnqwBeKPe51o5ry0T3Ckg/view?usp=drive_link(x * torch.sigmoid(x) ) 請問你們是有轉換成功的嗎? 如果有可以提供轉換的腳本做測試嗎? 後來提供…
-
@Andy Hsieh HI Andy 連結是將上面提供的模型替換activation function 為torch.sigmoid(x) 的 onnx 模型 https://drive.google.com/file/d/1WykBvQ0yZFbaVfw1C4dkOjkY-rytdtVy/view?usp=sharing 該onnx模型有確認是能夠成功轉換成nef的 再麻煩幫忙確認為何x * torch.s…
-
@Maria Chen Hi, Maria I tested your example for multiple chip which successes most of time. But I still get this error when load model first time. (Because your skip the disconnect chips, so I print the connect status) https://www.kneron.com/forum/…
-
@Maria Chen Hi Maria, I get the same results after change sleep_secs(1 sec, 5 secs, 10 secs) and replace init_device() with connect_devices(). After reset device, I check with scan_device() I found that I lost the chips(262) and chips(134) "is…
-
@Maria Chen Hi Maria, Does this test solve the "KP_ERROR_DEVICE_NOT_EXIST_10" error or just a test? I'm following your test steps below. https://www.kneron.com/forum/uploads/046/ETPV167CLPFA.png After reset, I'm losing one of chips. But t…
-
@Albert Chung Hi Andy 我上面提供的onnx模型已經是切一半的模型了,但還是無法轉換 目前觀察到的點主要是activation function如果是x * torch.sigmoid(x) 就無法轉換,需調整輸入大小 若改成torch.sigmoid(x)就可以轉換,且不須調整輸入大小(384*640) 我疑惑的點是只…
-
Hi Maria, I used the test code below. sometime load model on two chips will failed Model is the yolov5 in res/models/KL720/YoloV5s_640_640_3 https://www.kneron.com/forum/uploads/638/BPOEYTB6KX4H.png sometime load model on two chips will failed, but …
-
我好像是遇到跟這篇一樣的問題 https://www.kneron.com/forum/discussion/comment/1314#Comment_1314 裡面的建議是把feature map調小 feature map調小會導致準確率下降 但相同的模型我可以在v0.19.0轉換 但v0.24.0卻會超過硬體限制 還是我有其他參數可以修改?
-
Hi Maria, Check if you can find both KL720 chips by using scan_devices() https://www.kneron.com/forum/uploads/242/0TMD9UWPA4NT.png Try uploading model with only one KL720 chip to see if it works (and try it with the other KL720 chip too, separately…
-
上述用到toolchain的是latest,可能有誤 剛剛重新下載v0.24.0的toolchain 好像是可以成功轉換bie 但仍會有下述error Failure for model "yolov7_opt/yolov7_opt" when running "kdp720/HW not support" 這個是測試裁剪過後的onnx(並不是完整模型) …
-
I'm using custom development board with KL720 soc. I'm installing the python package of KneronPLUS and KneronPLUS YOLO. Your examples are for one chips. But I encounter this problme only when I'm loading model on two chips.
-
I found that when I execute the kp.core.load_model_from_file, the kl720 will disconnect and reconnect. The message below. [166578.771407] usb 2-1.2: USB disconnect, device number 84 [166578.975077] usb 2-1.1: USB disconnect, device number 83 [166580…
-
OS is busybox on armv7l GNU/Linux. Both chips are KL720 chips. Model is the yolov5 in res/models/KL720/YoloV5s_640_640_3 I just extract the sample code which only initialize device and load model below. The usb_port_ids are all correct. import os im…