requests.exceptions.HTTPError: 400 Client Error: BAD REQUEST for url: https://blobconverter.luxonis.com/compile?version=2021.4&no_cache=False

[NiloofarSabouri]

Hi
I am facing with this error to run my own dataset on an OAK.
Here is the error I got:
requests.exceptions.HTTPError: 400 Client Error: BAD REQUEST for url: https://blobconverter.luxonis.com/compile?version=2021.4&no_cache=False


I used yolov7.
Thanks.

[Erol444]

Hi @NiloofarSabouri ,
Seems like a problem with the model.. Could you share the json/bin/xml of your model, so we can repro/debug locally?

[NiloofarSabouri]

Hi @NiloofarSabouri , Seems like a problem with the model.. Could you share the json/bin/xml of your model, so we can repro/debug locally?

Hi, I have a folder named model and is empty.
when I used the luxions tool ,it gave me a zipped folder named result.
Here is result folder:

click the link all you want is right there.
https://drive.google.com/file/d/14o6ljo1wAsMHdXmjG0oqBPIg26ERq3kV/view?usp=drive_web
Thanks

[NiloofarSabouri]

Hi, thanks for your reply: here is my model: result.zip <https://drive.google.com/file/d/14o6ljo1wAsMHdXmjG0oqBPIg26ERq3kV/view?usp=drive_web> I have a folder named model and it is empty. When I used the luxions tool to convert my weight, it gave me a folder named result.zip as I sent it to you as a google drive link. but to run on my own oak I got those errors and pasted it in github. sincerely yours N.Sabouri

…

On Thu, Oct 5, 2023 at 5:43 PM Erol444 ***@***.***> wrote: Hi @NiloofarSabouri <https://github.com/NiloofarSabouri> , Seems like a problem with the model.. Could you share the json/bin/xml of your model, so we can repro/debug locally? — Reply to this email directly, view it on GitHub <#903 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AYFTHSD2AU23MPOWOC5K4QDX5257JAVCNFSM6AAAAAA5UI3MHGVHI2DSMVQWIX3LMV43OSLTON2WKQ3PNVWWK3TUHMYTONBYHE4DONZUGA> . You are receiving this because you were mentioned.Message ID: ***@***.***>

[Erol444]

@NiloofarSabouri I requested access on gdrive.

[NiloofarSabouri]

@NiloofarSabouri I requested access on gdrive.

Hi.
Now you have an access.

[Erol444]

Hi @NiloofarSabouri ,
Works for me. I just renamed folder to aaaa and all files to best11 so there wouldn't be any clash (also inside best11.json, change bin/xml paths).

[Erol444]

Do you have latest blobconverter installed? I have 1.4.2 version

[NiloofarSabouri]

Do you have latest blobconverter installed? I have 1.4.2 version

I used online blob convertor. (https://colab.research.google.com/corgiredirector?site=http%3A%2F%2Ftools.luxonis.com)
add my best.pt and select yolov7 (detection only) and it gave a .zip folder.
now it is running on my device...
but there is a problem sir...
it does not detect accurately..
let me send a picture..

[NiloofarSabouri]

How can I make it more accurate?
by the way I checked my result.zip with the COLAB tutorial ...
I don't have yolov7.json file , but it the Jupiter notebook that man I think Erick had 6 files in a zip folder..
Mode LastWriteTime Length Name

---

-a---- 12/27/2022 10:53 PM 28 .gitignore
-a---- 12/28/2022 3:34 PM 18074356 yolov6n-simplified.onnx
-a---- 12/28/2022 3:35 PM 8634878 yolov6n.bin
-a---- 12/28/2022 3:35 PM 2340 yolov6n.json
-a---- 12/28/2022 3:35 PM 163855 yolov6n.xml
-a---- 12/28/2022 3:35 PM 8691456 yolov6n_openvino_2021.4_6shave.blob

instead I have the best.json file.
but now the problem is OAK should detect more accurate...
Thanks...

[Erol444]

@NiloofarSabouri that shouldn't matter, you only need json/bin/xml. Regarding accuracy - I'd suggest larger dataset and further training, I don't think there's something wrong with OAK.

[NiloofarSabouri]

@NiloofarSabouri that shouldn't matter, you only need json/bin/xml. Regarding accuracy - I'd suggest larger dataset and further training, I don't think there's something wrong with OAK.

Thanks for your great kindness
I used 114 images to train.
Actually I want to detect 3 classes.
How many images do I need for this custom detection?

[Erol444]

It depends on the thing you'd want to detect usually. Do you get good results inside notebooks / when validating against your dataset (eg. 1/5 of dataset for validation)?

[NiloofarSabouri]

It depends on the thing you'd want to detect usually. Do you get good results inside notebooks / when validating against your dataset (eg. 1/5 of dataset for validation)?

Eroll
Actually I don't know how to make a good dataset.
And I have problems in training. First I wanted to train my dataset by yolo3 then I realized that colab has ended its supports.
so I decided to trian by Yolov7. In the tutorial I discovered they used VOC data set and then converted to YOLO format. My dataset is on Yolo format so I ignored the cells relating VOC and conversion to YOLO.
I realized that I have to make a folder named dataset and subdivided in to 2 folders which their names should be train and test and each folder should have two inner folders (images and labels).
Then I realized I have to make another folder , that should be val (for validating).
Now you said that I have to pick up 1/5 my data set for validation.
Can you explain more about what preprocessing methods do I need?

sorry I don't understand what do you mean (Do you get good results inside notebooks / when validating against your dataset (eg. 1/5 of dataset for validation)?

[Erol444]

Yes, that's correct. If you aren't exactly sure about the training part, perhaps Roboflow (or similar platforms) would be suggested.

[NiloofarSabouri]

Yes, that's correct. If you aren't exactly sure about the training part, perhaps Roboflow (or similar platforms) would be suggested.

Erol
Thanks .
let me increase my dataset and make you inform.