329 - What is Detectron2? An introduction.

This video provides an introduction to Detectron2 in python using pre-trained models for instance and panoptic segmentation.
Code generated in the video can be downloaded from here: github.com/bnsreenu/python_fo...
All other code:
github.com/bnsreenu/python_fo...
Detectron2 repo: github.com/facebookresearch/d...
What is Detectron2?
An open-source object detection and segmentation framework developed by Facebook AI Research.​
Built on top of PyTorch and provides a unified API for a variety of tasks, including object detection, instance segmentation, and panoptic segmentation.​
Designed to be flexible and easy-to-use, it puts a focus on enabling rapid research.​
It includes high-quality implementations of state-of-the-art algorithms like Mask R-CNN, RetinaNet, and DensePose.​
It includes a Model Zoo with models for object detection, instance segmentation, and more. ​

[@ajay0909]

I have done a lot of annotations for a project which took at least 3 hours per image. But with this detectron2 a lot of time and energy is saved. Thank you sir for detailed tutorial.

[@rahatkibriabhuiyan2426]

That's nice. Would I ask you which annotation tool you have used?

[@KalanaNethsara]

glad to see you again Mr. Sreeni, Love from SriLanka :)

[@yogidwitama2480]

Thank you, sir. I can't wait for the tutorial on using custom dataset. 😁

[@DigitalSreeni]

Working on it :)

[@yogidwitama2480]

@@DigitalSreeni 🔥🔥

[@rahatkibriabhuiyan2426]

@@DigitalSreeni
It is very delightful to hear that you're working with the custom dataset training.
I would request you to plz look into the matter to plot validation loss graph. I tried a lot of time but can not sort it out. With me tensor board I can plot training loss but not the validation loss. I have also search in the internet many people talk about the hook function or other things but those didn't work to me.
Thanks in advance.

[@pushkarkopparla260]

Thank you for sharing your code so freely!

[@DigitalSreeni]

Of course!! Thank you.

[@madeleinedawson8539]

Really helpful and I appreciate all these excellent videos!

[@hakankosebas2085]

please cover optimization problems, you explain topics very well

[@robosergTV]

Huggingface has all models that Detectron2 supports. Some models of Detectron2 (i.e. Mask2Former) can not be exported to ONNX or Torchscript for fast inference. HF does not have this issue.
Any reason you prefer Detectron2 over HF?

[@DigitalSreeni]

Detectron2 is known for its strong support for computer vision tasks, especially in object detection and segmentation. Hugging Face has gained popularity for its comprehensive model hub, which includes models for various natural language processing (NLP) and computer vision tasks. If your primary focus is on models that are readily exportable to ONNX or Torchscript for efficient deployment, Hugging Face might be a preferred choice. Basically the selection between Detectron2 and Hugging Face depends on your specific project requirements and the type of models you intend to use.

[@bibhutibaibhavbora8770]

Great video sir

[@eranfeit]

Thank you for the tutorial

[@DigitalSreeni]

You’re welcome 😊

[@shantilalzanwar8687]

How can we do basic math on detected region, if overlap is there, calculate area etc? Thanks for great content

[@lakayhukay]

how can i locate caves and tunnels sir? what can i can use

[@cplusplus-python]

Thanks PRof.
wondering about Diffusion + GAN models. Would be great if you could make it. Thanks again!

[@Tabelukhan]

Great Sir can you please make a video on Active learning would be grateful to you.!!

[@jayakrishna7995]

Great ... but sir please continue with microscopic datasets........

[@tamerzohdy8589]

Thanks a lot for these amazing videos.
Would you please make us some tutorials on Yolo algorithm

[@flloyd]

Thanks!

[@DigitalSreeni]

Thank you very much.

[@dounia4109]

Hello sir please I need the code for unmasking of masked face by gan

[@manoschaniotakis3328]

mmdetection is superior in code support and features / models, no?

[@user-nm1bi6xc6f]

Thank you for your videos. Just a question: is there any reason you have not talked at all about YOLO since it is actually state of the art in the area of image processing, object detection, segementation etc.. ?

[@DigitalSreeni]

I do not like YOLO for a very simple reason - I work with scientific images and in almost all cases I need masks as output and YOLO cannot do that. YOLO is fast and light weight but I think it is not for scientific images.

[@user-nm1bi6xc6f]

makes sense. Thank you for your reply @@DigitalSreeni 🙏

[@tamerzohdy8589]

Yolo algorithm can work with contours (polygons) in segmentation these polygons can be converted to binary masks. so why don't we try it?

[@michaelcollins399]

@@DigitalSreeni yolov8 outputs masks as well.

[@ShadowD2C]

@@DigitalSreenias others have said YOLOv8 outputs masks as well

[@loocalbertlin6258]

謝謝！

[@DigitalSreeni]

Thank you very much.

[@user-me7mo1iw2x]

Isn't it already outdated model? Sounds like it does not use VIT.

[@madeleinedawson8539]

Thanks!

[@DigitalSreeni]

Thank you very much.