I just want to know that it's can change or not
like this : https://github.com/BYLi-AI/poly-yolo
Thank you

The above issue occurred when I used Google Colab to transform the model，Excuse me, model yolov4_ Tiny_ Where to obtain voc.onnx？

Hi,

I encountered my Mac sometimes hang while running YoloV4Barracuda.
I am not sure what exactly make this happens because I run YoloV4 with other my logics to visualize the result but do you have any idea or any experience similar to this?
So far, I got this issue on MacbookPro16 2021 M1 Max with Unity Editor 2021.3.15f1 or 2021.3.26f1.
No issue found on Windows10 with Geforce 1080 running the same project.

I know some people reports hang issue on M1 Mac in UnityForum but am not sure my issue is the same one because the project works without problems when Yolo logic is turned off.

Any idea or tips would be helpful!

you say “This model was converted into ONNX by PINTO0309 (Katsuya Hyodo)”，Which project should I to choose specifically？

Is it possible custom training the model?

Hi,
Please, once imported the project how to launch the model to see the results?
Thanks in advance.

When I click play button in every Hierarchy, everything is great.
As I open the Marker IndirectDraw or Visualizer.cs these scripts, then tell me they can't find "YoloV4Tiny "
Did I forget to do something or missing some step? I'm so confused.
my English is not good, but I try my best to describe my problem and situation.
May you help me in your free time? Thanks!!

こんにちは。あなたのGethubを通じてたくさんの勉強をすることができました。 まず、お礼を申し上げます。
一つ質問がありまして、コメントを残させていただきます。

もし他の経路を通じて学習したonnxファイルをあなたのプロジェクトに使えるのか知りたいです。
私は一般的なPASCALVOCファイルの他に私が作ったデータモデルを適用させて該当モデルを修正したいです。

時間を割いて読んでくださってありがとうございます。

Hello, I was able to study a lot through your git hub. First of all, I would like to thank you.
I have a question, so I'm writing to you.

I wonder if I can use the .onnx file that I learned through another route for your project.
In addition to the general PASCALVOC file, I would like to modify the data model that I made by applying it.

Thank you for taking the time to read.

Overview

Hi,

I would like to modify Postprocess1.compute and Postprocess2.compute like the reference repository (https://github.com/bubbliiiing/yolov4-tiny-keras) and the general implementation method.
Additionally, I would like to add some features.

So I'm considering creating a pull request. I have committed all my changes to my forked repository and am ready to submit a pull request. Below is a summary of the changes. Does this repository accept pull requests?

Thank you.

Postprocess1.compute

Modifications

• Change variable names to more descriptive ones.
• Change the method of calculating scores to the product of confidence of the box and the predicted probability of the class: sigmoid function shall be applied to each.

Postprocess2.compute

Modifications

• Change variable names to more descriptive ones.
• Ensure that it is executed for each class.
• Sort the bounding boxes in order of scores before executing NMS.

Added functionalities

• Implementation of Soft-NMS, an evolution of NMS.
• Allow selection between NMS and Soft-NMS to be executed.

https://openaccess.thecvf.com/content_ICCV_2017/papers/Bodla_Soft-NMS_--_Improving_ICCV_2017_paper.pdf

Miscellaneous

According to the changes in the above files, some additional files need to be modified as well. We will enable changing hyperparameters related to NMS from the Inspector in the Scene.

When I use the Yolov4 tiny keras project to transform the model, I always cannot use it after importing Unity

when detection end we can see some object detection results like a game object in the real scene, not only for realtime detection

Hello, I was wondering why the input Image needs to be vertically flipped in Preprocess.compute?

训练出的权重文件是怎么转换为onnx格式的？

Hello, do you think v7 version can run through Barracuda?
Thanks