Відео

Thank you!

Indeed :)

Hi, I am always open to improvements. If you think you can improve the code, please do not hesitate to make a PR

Hi, thank you for your question. The results I show at the beginning of the video are without it.To me these are decent results although they would be better with the hierarchical volume sampling strategy. I think I will make a video about it in the near future :)

Thank you so so much!

Thank you!

Thank you so much!

You are absolutely right, thank you for the catch! This function was indeed borrowed from tensorflow. Feel free to make a pull request. Otherwise, I will update it myself.

Thank you! I show the results at the beginning of the video, and in the README -- link in the description

I will do my best to release it soon!

also, lots of echo

Hi @SleekGreek and @nickfratto2439, thank you for letting me know! I am redesigning my audio capture and hopefully it will be better in my next videos

Thank you!

Tank you! I am using Sublime Text

Yeah, I no longer use AI voice now

Thank you! This is planned! Should be released in the coming months

Thank you! Will do

Hi @anhtth2207, thank you for your question. Do you mean the sublime text theme? If yes, this is the default theme

My pleasure :)

Thank you so much!

Thank you! :)

Thank you for your question. The learned NeRF representation is a 3D model of the object. The most commonly used approach to obtain another representation (e.g. mesh) is to do a 3D to 3D conversion using algorithms such as Marching Cubes. Another possible approach, more closely related to what you suggest, is to use the NeRF representation to generate more views -- and potentially depths -- so that they can be fed to an algorithm such as TSDF (truncated signed distance function) Fusion.

Thank you so much! :)

Hi, thank you for your comment. I plan to release a video about implementing it from scratch in the near future.

Thankyou so much! Eagerly waiting!

Thank you for your comment! Yes, this is something I am working one. This will be one of my next videos when the code is ready

@@papersin100linesofcode This is truly amazing! Looking forward to a great video

Hi, thank you for your question. Unfortunately, not in high quality. We discuss the ray marching algorithm and use it to extract a mesh from NeRF. However, the mesh is not high quality, and does not possess colours. If you want a coarse mesh, that is fine, but if you have high expectations on the quality of the mesh, and need colours, then you would need more advances algorithms than the ones used in the course.

Glad to hear it, thank you! :)

Hi @OgJail, thank you for your interest. Yes, the email from the description is great

@@papersin100linesofcode Hello, it has been about two weeks since I’ve sent an email asking for more information about this opportunity. Is it still open?

@@papersin100linesofcodeI had emailed you two weeks ago but have not yet received a response. Is the position still open?

Thank you for your question. Yes, that is coming! :)

@@papersin100linesofcode looking forward to it :)

Thank you for your question and excuse me for the delayed answer. Are you talking about the sublime text color theme? If yes, to the best of my knowledge, these are the default colors.

Thank you!

Thank you!

Thank you so much!

Thank you! About 24 hours

More NeRF soon :) By updated in real time, do you mean rendered? If so, you might want to have a look at Volinga.ai.

For what i want to do, I need to dynamicly build a nerf. As new images are being streamed in it keeps updating the nerf kinda like a layzer point cloud. But i'm Not sure if that's posible @@papersin100linesofcode

Thank you!

Hi, thank you for your great comment! 1) I should have used a P5000 or RTX5000. 2) I am not sure I understand which line you are referring to?

Thank you so much!

Hi, thank you for your comment! I did not add the total variation regularization nor the other regularization proposed in the paper, for simplicity. I would be interested to see the impact on the synthetic data though :)

Hi, thank you for your question. This is because we concatenate the position to the positional encoding. This is not mentioned in the paper, but done in their implementation.

@@papersin100linesofcode ow. I understand. Thanks a lot

Hi, thank you for your comment. I am planning to add a video about it. I hope I can release it in the near future

@@papersin100linesofcode I subscribed, thank you

I am glad to hear that you like them! Thank you so much for your comment!

Excuse me for the delayed answer. This is second order because we compute the second order gradients.

@@papersin100linesofcode hey, thanks for the video. Where exactly are the second order gradients calculated? Wouldn't this necessitate create_graph=True in the inner_loop ?

​@@thomaswohrle1623 thank you for your question. It is computed on line 75 where the loss depends on theta_prime. For create_graph=True, this is a great question that I would need to investigate.

Excellent explanation, thanks a ton!

Hi, thank you for your question. Maybe you confuse weights and hidden units? All the weights are initialized randomly. Then, the hidden units of the first layer are computer from the weights of the first layer together with the input. Then, those hidden units are used to compute the hidden units of the following layers. I hope it is clear :) Let me know otherwise

@@papersin100linesofcode I know how to compute inputs and outputs as you go on to the next layers; but what weights and biases are used to compute the layers after that? For layer 1, you set them yourself, randomly. I don't see an equation to compute weights for all the layers after that. Are they set to random as well in the initial run, changed later by back optimization Thanks.

@@deeber35 Yes exactly! They are initialized randomly when each layer is iniitalized, and then, updated with backpropagation

I hope you liked the video :)

Thanks!

Hi, thank you for your question. Do you mean the camera to world matrix (c2w)? If so, yes, and actually the direction is already computed from it most of the time. The direction is computed from the camera, using its 3x3 c2w matrix

@@papersin100linesofcode yea can you please tell the formula that is used to get them?

@@aditya-bl5xh you may be interested in this script github.com/kwea123/nerf_pl/blob/master/datasets/ray_utils.py. I will soon make a video about it

@@papersin100linesofcode thanks! Appreciated

Hi, I show the results at the beginning of the video so that you know what results you can expect from the code of the video. I do not run the code because training is taking quite some time.

Hi@@papersin100linesofcode , Thank you for the reply. I was wondering if it is possible to create a video for NeRF applications on smartphone images. Anyway, thank you for your amazing content.

@@zeynolabedinsoleymani4591 yes this is planned! I will show how to reconstruct real-world scenes from phone cameras. I hope it is released in the coming weeks

Thank you for your question! For now it was hardcoded so that the size is the same as h, but it would be best to automate it.

Thank you for your comment! I will increase it in my next videos