GIoU vs DIoU vs CIoU | Losses | Essentials of Object Detection
HTML-код
- Опубликовано: 9 сен 2024
- This tutorial provides an in-depth and visual explanation of the three Bounding Box loss functions.
Other than the loss functions you would be able to learn about computing per sample gradients using the new Pytorch API.
Resources:
Colab notebook
colab.research...
Repo with results of training using different loss functions
github.com/ksa...
DIoU repo with matlab code and author's comment:
github.com/Zzh...
github.com/Zzh...
Just amazing. Could not have explained it any better. Thank you so much.
🙏
Thank you for these very clear and visually efficient explanations. I'll make sure to use these concepts in my PhD work !
good video glad you’re back, hoping to see initiating the training loop video soon
🙏
Great explanation. Thanks for sharing the video
🙏
Great content with a lot of optimization sir, hoping to see a videos related to vision transformers for object detection
🙏
Great content, keep it up. Thanks to these videos, my study of object detection is going much smoother.
🙏
It's really amazing job here. Thank you so much
🙏
Sir please continue posting video it's really helpful
🙏
Thanks for sharing -- thanks :)
🙏
I don’t comment that often on RUclips. I just came across your channel and I see you put a lot of effort in the content and the presentations as well. Your intuition is well aligned with mine in many videos . Good work ! Keep it up. Question: what software do you use for presentation (Manim, Adobe After Effects , etc ). The quality is good.
🙏 I use manim for very specific animations but 99% of time it is just PowerPoint. No after effects etc.
Impressive !. The equations are done in Manim or PowerPoint ? Last I recall the equations by PowerPoint equation editor don’t look that great.
@abdelwahedkhamiss in this tutorial, I have used manim for animation and equations. In others, I used latex. Yes, latex editor in PowerPoint is not well done at all.
I created a dataset of many million crop (mask and boundings box) to train mask-rcnn. It would be fun to see if mask-rcnn using giou would have an impact.
👍 😊
At 16:30, could you please explain what exactly did the user Zzh-tju comment? The first couple of sentences of his comment is somewhat understandable, but then it was just confusing. Could you please elaborate on that? Please be as detailed as possible. Thank you in advance!
When you do optimization using pytorch, auto differentiation (pytorch.org/tutorials/beginner/basics/autogradqs_tutorial.html) is used. Zzh-tju is suggesting that the gradients computed by the auto-diff are not same as one would compute by hand (in his case done in c and matlab) for some corner cases. These corner cases seem to influence the outcomes.
Hi great video! I would love to ask how did you manage to make the visualization of those convergence? I'm working on a similar topic, and I would love to try it on my modified loss
I used manim to generate the animation. www.manim.community
@@KapilSachdeva never heard about it but I just installed it.. if it possible, would you provide me your code pls?