How to Use MMDetection | Train RTMDet on a Custom Dataset

Рет қаралды 14,565

Күн бұрын

Dive into the world of computer vision with this comprehensive tutorial on training the RTMDet model using the renowned MMDetection library. Whether you're just starting out or looking to refine your skills, this guide offers a deep dive into the OpenMMLab ecosystem, hands-on installation steps, and practical insights into training on custom datasets.
Chapters:
- 00:00 Introduction
- 00:29 What is MMDetection and RTMDet
- 01:24 OpenMMLab Libraries Installation
- 04:54 Inference with Pre-trained COCO Model
- 08:14 Downloading a Dataset from Roboflow Universe
- 10:02 Preparing Custom MMDetection Configuration File
- 11:30 Train RTMDet and Analyze the Metrics
- 13:08 Evaluating the RTMDet Model with Supervision
- 14:24 Conclusions
Resources:
- 🌏 Roboflow: roboflow.com
- 📚 Roboflow Notebooks Repository: github.com/roboflow/notebooks
- 🌌 Roboflow Universe: universe.roboflow.com
- 📓 RTMDet Object Detection Google Colab: colab.research.google.com/git...
- 🗞️ RTMDet: An Empirical Study of Designing Real-Time Object Detectors Arxiv paper: arxiv.org/pdf/2212.07784.pdf
- 🗞️How to Train RTMDet on a Custom Dataset blog post: blog.roboflow.com/how-to-trai...
- 💻 MMDetection GitHub repository: github.com/open-mmlab/mmdetec...
- 🖼️ Double Twelve Dominoes object detection dataset: universe.roboflow.com/pip-tra...
Remember to like, comment, and subscribe for more content on AI, computer vision, and the latest technological breakthroughs! 🚀
Stay updated with the projects I'm working on at github.com/roboflow and github.com/SkalskiP! ⭐

Пікірлер: 57

@JemiloII Ай бұрын

The fact that you care about licensing that helps us all out, instant subscribe. Pretty tired of seeing AGPL Licensed code/models being used over and over.

@suphotnarapong355 10 ай бұрын

Thank you so much for sharing the knowledge I've been searching for. Your willingness to share this information has been incredibly helpful to me, and I truly appreciate it. It's made a significant difference in my understanding and learning. Thanks again!

@user-yl3fl8bs1b 11 ай бұрын

You're welcome for the helpful advice. I will try using it in my project.

@Roboflow 11 ай бұрын

Awesome! Please, let me know if you like it. And btw, was the LICENSE important factor for you?

@hartmannr76 10 ай бұрын

Hey, cool! You're using my domino dataset :) Thrilled to see it being used in something like this!!

@Roboflow 10 ай бұрын

So awesome! Shame I couldn’t contact you before. I’d love to test the model on the video input.

@Lena-qk7qq 4 ай бұрын

Awesome video, thanks!

@thanachai.j 11 ай бұрын

Thank for teaching .❤

@Roboflow 11 ай бұрын

My pleasure!

@abenezeryonathan3971 8 ай бұрын

Thank you so much for this tutorial. I wanted to implement tracking but I am having problem with MMTracker and MMYOLO. Can you do MMDetection and MMTracker model training with custom dataset in the future.... Please 🙏

@user-jj8qz2vy7o 8 ай бұрын

I've been trying to get mmdet working in a google colab project, and I keep running into an error where it fails to build a wheel for mmcv. The project from the video appears to have been running fine a few months back, but when I am running the code now, it gets the error I've been getting. Is anyone getting the same error? Does anyone have any insight or fixes?

@seanolivieri4829 8 ай бұрын

Me too: ERROR: Failed building wheel for mmcv ERROR: Could not build wheels for mmcv, which is required to install pyproject.toml-based projects ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. yfinance 0.2.31 requires requests>=2.31, but you have requests 2.28.2 which is incompatible. ModuleNotFoundError: No module named 'mmdet'

@riley8705 7 ай бұрын

Has anyone found a solution to this

@illyaaverchenko2263 3 ай бұрын

I solved by installing torch==2.0.1 and some other stuff before installing all the mmdetection stuff. !pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 torchdata==0.6.1 torchtext==0.15.2 fastapi kaleido python-multipart uvicorn lida torchdata

@zeeshanrizvi3350 Ай бұрын

Interestingly, I just solved this error a while ago. Try changing the version of torch to 2.0.0 and then run the command to install mmcv again. Hope it helps!

@murphy1162 11 ай бұрын

Great job!!! I just need it

@Roboflow 11 ай бұрын

We love to hear that! 🔥

@SkalskiP 11 ай бұрын

Thank you! 🙏

@tektronix475 11 ай бұрын

Does it have image anomaly detection?

@Roboflow 11 ай бұрын

@@tektronix475 when you say anomaly, do you mean fault detection?

@murphy1162 10 ай бұрын

@@Roboflowactually i am waiting for the latest SOTA swimTransformer for objectdetection!

@manujkumarjoshi9342 10 ай бұрын

Beautiful

@Roboflow 10 ай бұрын

Thank you!

@top-5-travel 7 ай бұрын

I have tried the google colab and it is not working. It doesn't install mmcv. Any ideas how to correct this?

@illyaaverchenko2263 3 ай бұрын

@hyperion6821 1 месяц назад I solved by installing torch==2.0.1 and some other stuff before installing all the mmdetection stuff. !pip install torch==2.0.1 torchvision==0.15.2 torchaudio==2.0.2 torchdata==0.6.1 torchtext==0.15.2 fastapi kaleido python-multipart uvicorn lida torchdata

@AdarshSingh-rm6er 2 ай бұрын

Hey thanks for the informative video. It was really intuitive and helped me to understand more about mmdetection model. I was using mmdetection model for face detection and getting error AttributeError: 'list' object has no attribute 'pred_instances' when i am training the model on the below line : detections = sv.Detections.from_mmdetection(results) Your help will be greatly appriciated.

@valliebear5706 Ай бұрын

Hi good tutorial, can I ask about the train loss and validation loss you showed the loss graph so which is the validation and train loss? is it possible to open the json and manually fetch it and use matplot? because if I use the mmdetection 3.* it does not prove the validation loss. thank you

@fredcdf75 9 ай бұрын

Thank you for this great tutorial! About the licenses, I think rtmdet is under Apache license only in mmdet but not in mmyolo (which is supposed to be a faster training version). In mmyolo rtmdet is under GPL v3 license. Please tell me if I am wrong I'd rather use mmyolo :)

@Roboflow 9 ай бұрын

This is really good question. Up until now thought that MMYolo offers only after training of MMDet model. So same model as MMDet. And only fast training code is under GPL.

@fredcdf75 9 ай бұрын

@@Roboflow Yes you're right the model should be the same but different training code. Anyway thanks again for taking the time to make these tutorials and replying to comments. I'm still a beginner and it helps a lot 👍

@gauravlochab9614 11 ай бұрын

very useful thanks! Can you make a tutorial on mm library as in whole

@Roboflow 11 ай бұрын

Hm... There are 30+ libraries in that ecosystem. That would probably be a very long video. What exactly would you like to learn?

@gauravlochab9614 10 ай бұрын

@@Roboflow mmsegmentation and mmdetection context -Given a detection problem and the need to select an appropriate model based on data health check in Roboflow, how should I proceed? For instance, when dealing with factors such as small boxes or high resolution within the data, how do I determine the suitable model in mmdetection and its necessary tuning? Your guidance is greatly appreciated. I'm an avid consumer of your content and a daily user of Roboflow for data preparation. Thank you.

@hysskm 9 ай бұрын

@Roboflow I want to know how to convert Our dataset's annotation format into coco format

@none-hr6zh 7 ай бұрын

how to make mmd datasets if I have already datasets in yolo and coco format.

@tryingtobeproductive 10 ай бұрын

7:35 I chuckled haha

@Roboflow 10 ай бұрын

Me too! Haha

@danielarmandovidalsoroa7787 3 ай бұрын

Hello, can we also deploy the custom-trained RTMdet model in roboflow? Or it does not work? Could you provide the code?

@amarnathreddysurapureddy9111 2 ай бұрын

Is it possible to train the Annotated Custom dataset using pytorch? I want train a model using pytorch but annotated images from roboflow/cvit/supervision. I have 600 images. I Annotated 600 iamges with different labels .I want classify dog, cat, rabbit, cow and mouse. Is it possible?

@lawrencerodriques7153 10 ай бұрын

Hey can you create a tutorial for smart Surveillance which detect human intrusion classify between authorised and unauthorised intruders and notify security person usign YOLO V8 please

@Roboflow 10 ай бұрын

How do I know which intrusion is authorized and unauthorized?

@lawrencerodriques7153 10 ай бұрын

@@Roboflow You will have to create authorised persons dataset and then verify like attendance system

@lawrencerodriques7153 10 ай бұрын

@@Roboflow please create it 🙏🙏

@Asadkhan-nz6qm 9 ай бұрын

Hello bro i have a question to the best of your knowledge which model is best for medical images?

@Roboflow 9 ай бұрын

Are we talking about segmentation or detection?

@diemagic5140 4 ай бұрын

Google Colab is not working please update it

@illyaaverchenko2263 3 ай бұрын

@marcelopedras88 5 ай бұрын

It no longer works on Colab. Can you solve this? ERROR: pip's dependency resolver does not currently take into account all the packages that are installed. This behaviour is the source of the following dependency conflicts. ipython 7.34.0 requires jedi>=0.16, which is not installed. lida 0.0.10 requires fastapi, which is not installed. lida 0.0.10 requires kaleido, which is not installed. lida 0.0.10 requires python-multipart, which is not installed. lida 0.0.10 requires uvicorn, which is not installed. cvxpy 1.3.3 requires setuptools>65.5.1, but you have setuptools 60.2.0 which is incompatible. google-colab 1.0.0 requires requests==2.31.0, but you have requests 2.28.2 which is incompatible. yfinance 0.2.36 requires requests>=2.31, but you have requests 2.28.2 which is incompatible.

@Roboflow 5 ай бұрын

Please create issue here and describe your problems: github.com/roboflow/notebooks/issues