UNetXST | Camera to Bird's Eye View | Perception for Self Driving Cars

Discussing a technique to convert images taken from different self driving cameras to a bird's eye view or top level view representation. The network makes use of a combination of UNet and Spatial Transformers.
Feel free to leave a comment or message me on Twitter/LinkedIn in case of any questions, doubts, suggestions or improvements.
Twitter: / mahnasakshay
LinkedIn: / sakshaymahna
Links
Dataset: www.kaggle.com...
Dataset Visualization: www.kaggle.com...
Notebook/Code: www.kaggle.com...
UNetXST Paper: arxiv.org/pdf/...
UNetXST Github Repository: github.com/ika...
UNet: towardsdatasci...
Image Transformations: kevinzakka.git...
Spatial Transformer Networks: kevinzakka.git...
Videos on Image Segmentation
---
Multi Task Attention Network: • Multi Task Attention N...
Fully Convolutional Network: • Fully Convolutional Ne...

[@loveneet_saini]

Wow!! Great video! Thank you for your time and clear explanation. Cheers!

[@user-st3dx8pd1o]

Awesome, thanks!

[@youstuffs5]

Great information 👍
I am figuring out how the same can be done using self-supervised method so that it can be applied to a different set of images for which bev labels are not available.

[@vaishalikharola8629]

But why we can use semnatlcly

[@josephgeorge495]

Is there any limit to the number of cameras?

[@granatapfel6661]

I would say no

[@mohammedatifkhan3834]

Please do path planning from a bird's eye view including explanation of algorithms.

[@RoboticswithSakshay]

Thank you! Really nice suggestion, I'll look into it!

[@RoboticswithSakshay]

Hello! You can checkout the latest video. Link: ruclips.net/video/nbaSzCnmyec/видео.html
Hope you like it.

[@jimj2683]

Why has nobody collected data from the real world? You could attach a pole with a downward-facing camera on top of a bunch of teslas and drive around collecting data. Then you have all the ground truth you need for making 360 parking cameras much better.

[@PranshuTople]

Hey Sakshay! Great video. I've sent you a mail. Would you please have a look at it! 😉

[@RoboticswithSakshay]

Hello Pranshu! Yes, I'll have a look.