Blobconverter SSL certificate error

AyushmanChoudhuri

I am trying to use the blobconverter api to convert my onnx model files to the .blob format using this line of code:

blob_path = blobconverter.from_onnx(model="yolov6n.onnx", data_type="FP16",shaves=6,version="2022.1")

However I am facing this a SSLCertVerification error:

SSLCertVerificationError Traceback (most recent call last)

/usr/local/lib/python3.11/dist-packages/urllib3/connectionpool.py in make_request(self, conn, method, url, body, headers, retries, timeout, chunked, response_conn, preload_content, decode_content, enforce_content_length)
463 try:
--> 464 self.validate_conn(conn)
465 except (SocketTimeout, BaseSSLError) as e:
20 frames

$$
SSLCertVerificationError: [SSL: CERTIFICATE_VERIFY_FAILED] certificate verify failed: certificate has expired (_ssl.c:1016)
$$

KlemenSkrlj

Hi @AyushmanChoudhuri ,
Looks like the SSL Certification didn't update automatically. I just updated it and you should be able to run it now - apologies for inconvinience.

Since you are using Blobconverter which is now deprecated I would also like to pitch you the usage of our Hub platform where you can convert the model through UI (documentation here) or if you prefer the python API through our modelconverter package (documentation here). If you specifically want to convert to .blob you have the checkbox under advanced settings. But I would recommend switching to DepthAIv3 if possible and then using the .superblob NNArchive because with that one you can convert once and then change shaves for each pipeline separately (instead of having multiple .blob files with different shaves).

Let me know if you have any additional questions.
Klemen

AyushmanChoudhuri

Hi Klemen

Thank you for your response. The blobconverter api works now.

I have been using the blobconverter api to generate .blob files for our system as the modelconverter api still outputs .superblob files even when I pass the superblob=False argument. I have attached the code snippet below:

from modelconverter import convert

converted_model = convert.RVC2( api_key=HUBAI_API_KEY,path="yolov6n.pt", name="YOLOv6 nano VOC", description_short="Trained YOLOv6 nano object detection model on VOC dataset.", yolo_version="yolov6r3", superblob=False, yolo_input_shape="640 640", yolo_class_names=labels, tasks=["OBJECT_DETECTION"], license_type="MIT", is_public=False)

We plan to move to the .superblob NNArchive format soon, but for the time being we would still need our models in .blob format.

KlemenSkrlj

I see. This is a bug on our side yes, we'll work on the fix for next modelconverter release. Thanks for the report.
Klemen

AyushmanChoudhuri

Hi Klemen

Thank you for the update. Looking forward to the bug fix.

Few more points:

I noticed that the https://tools.luxonis.com/ shows an axios network error at the moment and does not allow the conversion of pytorch files. Would be great if this was resolved until we can use the modelconverter api to directly generate blob files 🙂
For the generation of blob files on the hub UI platform, a onnx model graph of the yolov6r3 model is needed in a certain format. Are there any scripts/tools available from luxonis which assist with this export. The standard model graph available from onnx export scripts from the yolov6 repository is different from the one generated by https://tools.luxonis.com/

KlemenSkrlj

I noticed that the https://tools.luxonis.com/ shows an axios network error at the moment

Are you able to share the .pt file (can be untrained model) with us so I can use it to test and fix? You can share them throgh google drive or wetransfer or anything else. I tried yolov6r3 weights right now and it worked but might me something specific with your weights.

For the generation of blob files on the hub UI platform, a onnx model graph of the yolov6r3 model is needed in a certain format

When converting YOLO models through Hub you should click the "Convert Model - YOLO" button. This allow you to upload pytorch file directly and it will first convert it to ONNX (using tools package) and then to desired platform (using modelconverter package). I hope this clears things up.