ultralytics / yolov5

YOLOv5 πŸš€ in PyTorch > ONNX > CoreML > TFLite
https://docs.ultralytics.com
GNU Affero General Public License v3.0
49.6k stars 16.1k forks source link

where is yolov5 paper? #1333

Closed eeric closed 3 years ago

eeric commented 3 years ago

❔Question

where is paper

Additional context

github-actions[bot] commented 3 years ago

Hello @eeric, thank you for your interest in our work! Please visit our Custom Training Tutorial to get started, and see our Jupyter Notebook Open In Colab, Docker Image, and Google Cloud Quickstart Guide for example environments.

If this is a bug report, please provide screenshots and minimum viable code to reproduce your issue, otherwise we can not help you.

If this is a custom model or data training question, please note Ultralytics does not provide free personal support. As a leader in vision ML and AI, we do offer professional consulting, from simple expert advice up to delivery of fully customized, end-to-end production solutions for our clients, such as:

For more information please visit https://www.ultralytics.com.

glenn-jocher commented 3 years ago

@eeric hi thanks for asking! We aim to publish a short paper to arxiv by the end of 2020. In the meantime please see https://doi.org/10.5281/zenodo.3908559 for DOI to cite this repo directory, and where you can also find bibtex and other exportable citation formats.

DOI

hujunchao commented 3 years ago

Looking forward to meeting YOLOv5 paper!

github-actions[bot] commented 3 years ago

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.

NavneetSajwan commented 3 years ago

Is the paper out yet?

selenecodes commented 3 years ago

Any updates on this?

skullyhoofd commented 3 years ago

This probably shouldn't be closed yet.

robertokcanale commented 3 years ago

Any updates?

vikxoxo commented 3 years ago

Any paper or detailed doc, illustrating the architecture

glenn-jocher commented 3 years ago

@vikxoxo @robertokcanale it's not as good as a paper, but you can visualize the architectures now after our Tensorboard fix in PR https://github.com/ultralytics/yolov5/pull/2758. Hopefully paper will follow soon.

This is a YOLOv5s model displayed in TensorBoard. You can see the Detect() layer merging the 3 layers into a single output for example, and everything appears to work and visualize correctly. You can get this by uncommenting the tb_writer.add_graph() lines 333 and 335 in train.py: https://github.com/ultralytics/yolov5/blob/0f395b3e3bccbc019ab3d1cbd41303a5b50dc0f0/train.py#L333-L335

Screenshot 2021-04-11 at 01 10 09
robertokcanale commented 3 years ago

@glenn-jocher Thank you, I'll try it out. This looks good enough for my Master Thesis! But as I might also plan a publication on this, a paper would be of great help, can't wait for it, hoping reviewers accept it smoothly!

Muhtadir commented 3 years ago

@robertokcanale same here sir.

cenkbircanoglu commented 3 years ago

+1

github-actions[bot] commented 3 years ago

πŸ‘‹ Hello, this issue has been automatically marked as stale because it has not had recent activity. Please note it will be closed if no further activity occurs.

Access additional YOLOv5 πŸš€ resources:

Access additional Ultralytics ⚑ resources:

Feel free to inform us of any other issues you discover or feature requests that come to mind in the future. Pull Requests (PRs) are also always welcomed!

Thank you for your contributions to YOLOv5 πŸš€ and Vision AI ⭐!

MaKaNu commented 3 years ago

It is 2021 and still no paper. With no paper this Work is pretty much useless from a scientific point of view.

bakirsw commented 3 years ago

I there any chance that it will be published this month?

ksasso1028 commented 3 years ago

a paper would be really nice. But the fact that you have the architecture in pure pytorch is very helpful. Anyone with enough DL exp could trace this arch out and break down the types of layers you are using. I wish every paper that came out had a Pytorch implementation

suryagutta commented 3 years ago

A paper would be beneficial!

rossGardiner commented 3 years ago

Will there be a paper for Yolov5 before the end of the year? In a research context, I can't really commit to using Yolov5 if there is no paper describing inner workings...

zxsitu commented 3 years ago

@rossGardiner @glenn-jocher Same feeling as you. I am also a bit worried about using yolov5 for my article at the moment, although it is an excellent engineering project, the lack of an introductory article is really hard to convince the reviewers of the journal.

Guemann-ui commented 2 years ago

Hey, is the Yolov5 paper out yet?

glenn-jocher commented 2 years ago

@besmaGuesmi we are targeting a paper release by the start of PyTorch Dev day, December 1st.

If the YOLOv5 paper is not published by then I will eat my hat.

Guemann-ui commented 2 years ago

@glenn-jocher sounds great! all the best.

Keiku commented 2 years ago

The next release may be less confusing if the paper and code are released at the same time. I'm looking forward to it.

Makki93 commented 2 years ago

@glenn-jocher any news about the paper yet? :) 1st of December is soon and I wanted to reference it in my thesis with the due date end of December. Also, I don't want to see you eating a hat.

muhammedakyuzlu commented 2 years ago

Today is the premise day @glenn-jocher

maxl-13 commented 2 years ago

everyone is still waiting :)

fazer125 commented 2 years ago

@glenn-jocher Where is paper? Will you eat your hat?

felixdittrich92 commented 2 years ago

@glenn-jocher any update ? :)

gkm0120 commented 2 years ago

@glenn-jocher Now on December 2, 2021, there are still no papers :)

johnmarkacala commented 2 years ago

@vikxoxo @robertokcanale it's not as good as a paper, but you can visualize the architectures now after our Tensorboard fix in PR #2758. Hopefully paper will follow soon.

This is a YOLOv5s model displayed in TensorBoard. You can see the Detect() layer merging the 3 layers into a single output for example, and everything appears to work and visualize correctly. You can get this by uncommenting the tb_writer.add_graph() lines 333 and 335 in train.py:

https://github.com/ultralytics/yolov5/blob/0f395b3e3bccbc019ab3d1cbd41303a5b50dc0f0/train.py#L333-L335

Screenshot 2021-04-11 at 01 10 09

I cant seem to find this in train.py anymore. Any other recommendations?

Makki93 commented 2 years ago

There is this paper: TPH-YOLOv5: Improved YOLOv5 Based on Transformer Prediction Head for Object Detection on Drone-captured Scenarios https://arxiv.org/pdf/2108.11539.pdf

As well as this thesis: from https://www.theseus.fi/bitstream/handle/10024/452552/Do_Thuan.pdf

Quote: " 5TH GENERATION OF YOLO A month after YOLOv4 was released, researcher Glenn and his team published a new version of the YOLO family, called YOLOv5 (Jocher, 2020). Glenn Jocher is a researcher and CEO of Ultralystics LLC. YOLO models were developed on a custom framework Darknet which is written mainly in C by Alexey Bochkovsky. Ultralystic is the company that converts previous versions of YOLO on one of the most famous frameworks in the field of deep learning, PyTorch which is written in the Python language. 4.1 Overview of YOLOv5 Besides, Glenn Jocher is also the inventor of the Mosaic data augmentation and acknowledged by Alexey Bochkovsky in the YOLOv4 paper (Bochkovskiy, et al., 2020). However, his YOLOv5 model caused lots of controversy in the computer vision community because of its name and improvements. Despite being released a month after YOLOv4, the start of research for YOLOv4 and YOLOv5 was quite close (March – April 2020). For avoiding collision, Glenn decided to name his version of YOLO, YOLOv5. Thus, basically, both researchers applied the state-of-the-art innovations in the field of computer vision at that time. That makes the architecture of YOLOv4 and YOLOv5 very similar and it makes many people dissatisfied with the name YOLOv5 (5th generation of YOLO) when it does not contain multiple outstanding improvements compared to the previous version YOLOv4. Besides, Glenn did not publish any paper for YOLOv5, causing more suspicions about YOLOv5. However, YOLOv5 possessed the advantages in engineering. YOLOv5 is written in Python programming language instead of C as in previous versions. That makes installation and integration on IoT devices easier. In addition, the PyTorch community is also larger than the Darknet community, which means that PyTorch will receive more contributions and growth potential in the future. Due to being written in 2 different languages on 2 different frameworks, comparing the performance between YOLOv4 and YOLOv5 is difficult to be accurate. But after a while, YOLOv5 37 has proved higher performance than YOLOv4 under certain circumstances and partly gained confidence in the computer vision community besides YOLOv4. 4.2 Notable differences – Adaptive anchor boxes As mentioned above, the YOLOv5 architecture has integrated the latest innovations similar to the YOLOv4 architecture, thus there are not many brilliant differences in theory. The author did not publish a detailed paper, but only launched a repository on Github and updates improvements there. By dissecting its structure code in file .yaml, the YOLOv5 model can be summarized as follows (Jocher, 2020):

Keiku commented 2 years ago

Where is the yolov5 paper written by "glenn-jocher"? This is the correct issue. Everyone is waiting for it because he commented on it himself.

dinhanhx commented 2 years ago

21 days till 2022, still no paper?

MaKaNu commented 2 years ago

@glenn-jocher I prepared it for you: Tk0uEdf9OPQuQFghgiBDaBm-J5hA3btAcVrZhckgUtQ

Sorry I had to do it ;)

Makki93 commented 2 years ago

I don't want to stir up hatred here, but I have the bad feeling that YOLOv5 just doesn't perform as well as wanted to release a paper with results in a general comparison. Or they just don't do anything majorly different to v4 so it would justify a scientific paper.

I understand if what I stated above is true, that it's a strategy to not release a paper yet. Ultralytics can't afford any bad reputation when they want to earn money in the future with YoloV5. But I don't get the point of making promises that aren't kept or even answered anymore.

They could just say they're only focussing on YoloV5 being a well (not superior) performing object detector which is super easy to integrate everywhere, as they already do. Nothing more.

Makki93 commented 2 years ago

Here's a benchmark image from the YOLOv4 author, published over pjreddie's darknet repository (YOLOv1-3), to underpin my statement:

image

source: https://github.com/pjreddie/darknet

Alzubair98 commented 2 years ago

21 days till 2022, still no paper?

only few hours now, still no paper

dinhanhx commented 2 years ago

21 days till 2022, still no paper?

only few hours now, still no paper

And it's already 2022 in Japan

ssooffiiaannee commented 2 years ago

No paper yet? how can we know the difference between yolov5s yolov5x yolov5l and others? I cannot run these models without knowing what they are doing, no need for a paper, just a small article. @glenn-jocher

darkC00kie commented 2 years ago

I would also be really interested in reading a paper about the architecture and yolo in general as I am writing my thesis on this subject.

@glenn-jocher

AyadAlmamary commented 2 years ago

Hi, while we are waiting for Yolo-v5 paper, I want to test my dataset on Yolo3 and v4, anyone has a Colab file to do the training on custom dataset. thanks

Guemann-ui commented 2 years ago

@AyadAlmamary https://medium.com/analytics-vidhya/train-a-custom-yolov4-object-detector-using-google-colab-61a659d4868

AyadAlmamary commented 2 years ago

@besmaGuesmi You are here also , Thanks a lot.

vikxoxo commented 2 years ago

You may also search for Roboflow tutorials for yolo v3 & 4

On Sun, Jan 23, 2022, 04:28 Ayad Almamary @.***> wrote:

Hi, while we are waiting for Yolo-v5 paper, I want to test my dataset on Yolo3 and v4, anyone has a Colab file to do the training on custom dataset. thanks

β€” Reply to this email directly, view it on GitHub https://github.com/ultralytics/yolov5/issues/1333#issuecomment-1019372766, or unsubscribe https://github.com/notifications/unsubscribe-auth/AJ6AANYPNOB5TLPYAPBY5FLUXMZBDANCNFSM4TQL7FVQ . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

You are receiving this because you were mentioned.Message ID: @.***>

AyadAlmamary commented 2 years ago

@glenn-jocher Hi, we are waiting for YOLO-v5 paper, we want to cite it in our researches paper, Any update about it? -Also, I want a way to contact with you, email or something else?

Vishal-rkavk commented 2 years ago

any update on YOLOV5 paper? @glenn-jocher

ssooffiiaannee commented 2 years ago

@AyadAlmamary his email : glenn.jocher@ultralytics.com, I wrote him, he has answered my questions except the one about the paper. If there is no paper, please just say it @glenn-jocher , I think avoiding all questions about the paper will lead to distrust.

BanzerUniKL commented 2 years ago

I actually have to use a different model for my master-thesis if there won't come a paper. please give us informations about the state; if there won't come it is fine, i just need to know

MarigoldJ commented 2 years ago

I just got here for searching YOLO paper, but I found that only v5 paper doesn't exist. :(