Yolov4 Model Compilation issue wrt change in Num_classes

[PavanTikotikar]

we referring to https://github.com/Xilinx/Vitis-AI-Tutorials/tree/master/Design_Tutorials/07-yolov4-tutorial. "When Num_classes is 16 changed in dpu_yolov4.prototxt. then model is not working properly." In our case we are doing transfer learning on model pre-trained on coco dataset. We are not getting any error or crashes. But when we deployed our model, it is not working. It is not detecting classes in runtime. But our model before conversion worked fine on CPU.

[jimheaton]

Are you using the Darknet to Caffe conversion flow, or using TensorFlow? Does the detection work properly when using the quantized mode on the CPU/GPU?

[PavanTikotikar]

Hi Jim, Tutorial has explained how to train yolov4 with coco dataset. We have followed the steps explained in the tutorial and trained the model with coco dataset. And we have converted the model to xmodel and deployed to Kria board. And this model is working fine.

But we have our own dataset with 16 different classes. So, we did transfer learning using earlier trained yolov4. And we have followed the same steps to covert the model to xmodel and deployed to Kria board. But this time it's not working. But the same model before conversion is working fine on CPU.

[jimheaton]

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the Vitis-AI Library sw application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points do not match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

[PavanTikotikar]

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards, Pavan Tikotikar Avnet India Pvt Ltd +91-7022819794

From: Jim Heaton @.> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.> Cc: Tikotikar, Pavan @.>; Author @.> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

— Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[PavanTikotikar]

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.>; Mayur Lathiya @.> Cc: Author @.>; Dharin Shah @.> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.> Cc: Tikotikar, Pavan @.>; Author @.> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

— Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[PavanTikotikar]

Hi Pavan,

We are still waiting for the reply. Can you please let us know what step we doing wrong? Is there anything wrong with the files that we have shared?

Regards, Mayur Lathiya

---

From: Mayur Lathiya @.> Sent: 29 March 2022 11:42 To: Tikotikar, Pavan @.>; Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.> Cc: Author @.>; Dharin Shah @.> Subject: Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.>; Mayur Lathiya @.> Cc: Author @.>; Dharin Shah @.> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.> Cc: Tikotikar, Pavan @.>; Author @.> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

— Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[PavanTikotikar]

Adding ++ Nutan

Hi Nutan,

Mayur have already shared supporting required file . We waiting for JIM's reply. Please check and let us know.

Thanks and Regards, Pavan Tikotikar Avnet India Pvt Ltd +91-7022819794

From: Mayur Lathiya @.> Sent: Wednesday, April 6, 2022 3:45 PM To: Tikotikar, Pavan @.>; Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.> Cc: Author @.>; Dharin Shah @.> Subject: [External] Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

We are still waiting for the reply. Can you please let us know what step we doing wrong? Is there anything wrong with the files that we have shared?

Regards, Mayur Lathiya

---

From: Mayur Lathiya @.**@.>> Sent: 29 March 2022 11:42 To: Tikotikar, Pavan @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.**@.>> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Mayur Lathiya @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.**@.>> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Tikotikar, Pavan @.**@.>>; Author @.**@.>> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

- Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[jimheaton]

I do not have any files.

[jimheaton]

The files are not large, maybe you can paste the contents in the comment.

[PavanTikotikar]

Adding ++ Aditya

FYI

From: Tikotikar, Pavan Sent: Thursday, April 7, 2022 10:13 AM To: Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.>; Jim Heaton @.> Cc: Author @.>; Dharin Shah @.>; Nutan Rajaram Ghatge @.>; Mayur Lathiya @.***> Subject: FW: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi JIM,

FYI. I am resharing required files. Please let us know. Did you receive or not..

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Thanks and Regards, Pavan Tikotikar Avnet India Pvt Ltd +91-7022819794

From: Mayur Lathiya @.**@.>> Sent: Tuesday, March 29, 2022 11:43 AM To: Tikotikar, Pavan @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: [External] Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.**@.>> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Mayur Lathiya @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.**@.>> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Tikotikar, Pavan @.**@.>>; Author @.**@.>> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

- Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[PavanTikotikar]

Hi Pavan,

I have shared the files with Jim.

NUTAN GHATGE Senior Product Application Engineer | AMD Adaptive and Embedded Computing Group (AECG)

Salarpuria Sattva Knowledge City, Hitec City, Hyderabad, Telangana 500081, India Facebookhttps://www.facebook.com/AMD | Twitterhttps://twitter.com/AMD | amd.comhttp://www.amd.com/ @. Xilinx is now part of AMDhttps://www.amd.com/en/corporate/xilinx-acquisition @.https://docs.xilinx.com/

From: Tikotikar, Pavan @.> Sent: Thursday, April 7, 2022 4:12 PM To: Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.>; Jim Heaton @.>; Aditya Sivagnanam @.> Cc: Author @.>; Dharin Shah @.>; Nutan Rajaram Ghatge @.>; Mayur Lathiya @.***> Subject: FW: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

CAUTION: This message has originated from an External Source. Please use proper judgment and caution when opening attachments, clicking links, or responding to this email.

Adding ++ Aditya

FYI

From: Tikotikar, Pavan Sent: Thursday, April 7, 2022 10:13 AM To: Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Jim Heaton @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>>; Nutan Rajaram Ghatge @.**@.>>; Mayur Lathiya @.**@.>> Subject: FW: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi JIM,

FYI. I am resharing required files. Please let us know. Did you receive or not..

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Thanks and Regards, Pavan Tikotikar Avnet India Pvt Ltd +91-7022819794

From: Mayur Lathiya @.**@.>> Sent: Tuesday, March 29, 2022 11:43 AM To: Tikotikar, Pavan @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: [External] Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.**@.>> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Mayur Lathiya @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.**@.>> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Tikotikar, Pavan @.**@.>>; Author @.**@.>> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

- Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[jimheaton]

I took a look at the .cfg and the .protxt files and they look correct.

Have you tried running a detection using the quantized model on your cpu or gpu host machine? Is the quantized model working as expected on the host?

Once you verify that the quantized model is working, then you can try running on your board and try debugging using Vart (see Ch5 of UG141).

[PavanTikotikar]

Hi Mayur,

Below Is the reply from Vitis AI tutorial team,

I took a look at the .cfg and the .protxt files and they look correct.

Have you tried running a detection using the quantized model on your cpu or gpu host machine? Is the quantized model working as expected on the host?

Once you verify that the quantized model is working, then you can try running on your board and try debugging using Vart (see Ch5 of UG1414).

Thanks and Regards, Pavan Tikotikar Avnet India Pvt Ltd +91-7022819794

From: Mayur Lathiya @.> Sent: Wednesday, April 6, 2022 3:45 PM To: Tikotikar, Pavan @.>; Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.> Cc: Author @.>; Dharin Shah @.> Subject: [External] Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

We are still waiting for the reply. Can you please let us know what step we doing wrong? Is there anything wrong with the files that we have shared?

Regards, Mayur Lathiya

---

From: Mayur Lathiya @.**@.>> Sent: 29 March 2022 11:42 To: Tikotikar, Pavan @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.**@.>> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Mayur Lathiya @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.**@.>> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Tikotikar, Pavan @.**@.>>; Author @.**@.>> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

- Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[PavanTikotikar]

Hi Mayur,

Any updates. Did check try running on board by debugging using Vart as mentioned in UG1414 (see Ch5 of UG1414).

Thanks and Regards, Pavan Tikotikar Avnet India Pvt Ltd +91-7022819794

From: Tikotikar, Pavan Sent: Friday, April 8, 2022 9:13 AM To: Mayur Lathiya @.>; Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.> Cc: Author @.>; Dharin Shah @.***> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Mayur,

Below Is the reply from Vitis AI tutorial team,

I took a look at the .cfg and the .protxt files and they look correct.

Have you tried running a detection using the quantized model on your cpu or gpu host machine? Is the quantized model working as expected on the host?

Once you verify that the quantized model is working, then you can try running on your board and try debugging using Vart (see Ch5 of UG1414).

Thanks and Regards, Pavan Tikotikar Avnet India Pvt Ltd +91-7022819794

From: Mayur Lathiya @.**@.>> Sent: Wednesday, April 6, 2022 3:45 PM To: Tikotikar, Pavan @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: [External] Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

We are still waiting for the reply. Can you please let us know what step we doing wrong? Is there anything wrong with the files that we have shared?

Regards, Mayur Lathiya

---

From: Mayur Lathiya @.**@.>> Sent: 29 March 2022 11:42 To: Tikotikar, Pavan @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.**@.>> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Mayur Lathiya @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.**@.>> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Tikotikar, Pavan @.**@.>>; Author @.**@.>> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

- Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[PavanTikotikar]

Hi Pavan,

Thank you for your support.

I haven't tried that as I have modified our existing dataset and started training the model from scratch without transfer learning. And this new model is seems to working fine. You can close the issue for now.

If we need more support we will contact you.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.> Sent: 14 April 2022 14:56 To: Mayur Lathiya @.>; Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.> Cc: Author @.>; Dharin Shah @.> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Mayur,

Any updates. Did check try running on board by debugging using Vart as mentioned in UG1414 (see Ch5 of UG1414).

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Tikotikar, Pavan Sent: Friday, April 8, 2022 9:13 AM To: Mayur Lathiya @.>; Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.> Cc: Author @.>; Dharin Shah @.***> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Mayur,

Below Is the reply from Vitis AI tutorial team,

I took a look at the .cfg and the .protxt files and they look correct.

Have you tried running a detection using the quantized model on your cpu or gpu host machine? Is the quantized model working as expected on the host?

Once you verify that the quantized model is working, then you can try running on your board and try debugging using Vart (see Ch5 of UG1414).

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Mayur Lathiya @.**@.>> Sent: Wednesday, April 6, 2022 3:45 PM To: Tikotikar, Pavan @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: [External] Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

We are still waiting for the reply.

Can you please let us know what step we doing wrong? Is there anything wrong with the files that we have shared?

Regards, Mayur Lathiya

---

From: Mayur Lathiya @.**@.>> Sent: 29 March 2022 11:42 To: Tikotikar, Pavan @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards,

Mayur Lathiya

---

From: Tikotikar, Pavan @.**@.>> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Mayur Lathiya @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.**@.>> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Tikotikar, Pavan @.**@.>>; Author @.**@.>> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

— Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>

[bhargavin1872008]

when running the requirements.txt of keras-yolov3-modelset -i 'm getting error for coremltools.it is showing like "couldn't find a version that satisfies the requirement tensorflow<=1.14 and tensorflow >=1.5(from tfcoremltools -r requirements.txt).(from version :2.2.0,2.2..1, 2.2.2, ...2.7.0rc0,2.7.0.rc1............) like this .can someone help me regarding this. Also ,i have a doubt .can we use ubuntu 20.04 ,cuda 11.7 ,cudnn 8.4.0 for this project. or have to use ubuntu 18.04,cuda 10.0 only which only works.please help me regarding this,i have less time in my hand.

[PavanTikotikar]

Hi JIM,

FYI. I am resharing required files. Please let us know. Did you receive or not..

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Thanks and Regards, Pavan Tikotikar Avnet India Pvt Ltd +91-7022819794

From: Mayur Lathiya @.> Sent: Tuesday, March 29, 2022 11:43 AM To: Tikotikar, Pavan @.>; Xilinx/Vitis-AI-Tutorials @.>; Xilinx/Vitis-AI-Tutorials @.> Cc: Author @.>; Dharin Shah @.> Subject: [External] Re: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

Hi Pavan,

I am attaching the files mentioned in the mail chain.

The issue we are facing is we are not getting prediction. Our custom trained Yolo model is giving correct prediction on CPU but after deploying it to Xilinx board it not giving any prediction. We are not getting crash or wrong prediction but it is not able predict.

Regards, Mayur Lathiya

---

From: Tikotikar, Pavan @.**@.>> Sent: 26 March 2022 10:34 To: Xilinx/Vitis-AI-Tutorials @.**@.>>; Xilinx/Vitis-AI-Tutorials @.**@.>>; Mayur Lathiya @.**@.>> Cc: Author @.**@.>>; Dharin Shah @.**@.>> Subject: RE: Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

++ Mayur,

Hi Mayur,

Kindly go through below mail chain. And Please share your darknet .cfg file and the .prototxt file you are using.

Thanks and Regards,

Pavan Tikotikar

Avnet India Pvt Ltd

+91-7022819794

From: Jim Heaton @.**@.>> Sent: Friday, March 25, 2022 10:44 PM To: Xilinx/Vitis-AI-Tutorials @.**@.>> Cc: Tikotikar, Pavan @.**@.>>; Author @.**@.>> Subject: [External] Re: [Xilinx/Vitis-AI-Tutorials] Yolov4 Model Compilation issue wrt change in Num_classes (Issue #58)

What does not working mean? Do you get incorrect predictions or does the application terminate. Are you using the Vitis AI Library sw example?

I have found that when the application does not run correctly on the hw, it is usually an issue with the .prototxt file on the board where the parameters such as number of classes or anchor points due match what the model was trained with in darknet.

Please share your darknet .cfg file and the .prototxt file you are using.

- Reply to this email directly, view it on GitHubhttps://github.com/Xilinx/Vitis-AI-Tutorials/issues/58#issuecomment-1079233334, or unsubscribehttps://github.com/notifications/unsubscribe-auth/AYG6Q6OXHALWS557TGSE3H3VBXX5PANCNFSM5QVWPT3Q. You are receiving this because you authored the thread.Message ID: @.**@.>>