Use PIFuHD to Create a 3D Model of Anyone from a Single Photo

Рет қаралды 127,653

Күн бұрын

March 24, 2023: Fixed! This Colab is old and was giving errors, I have updated it to import the correct libraries. Had good results with a standard stock photo of a person standing. Updated Colab Notebook: colab.research.google.com/git...
Result from my tests of the fixed colab here: • Result from PIFuHD, Ma...
April 24, 2022: Google just released PHORHUM phorhum.github.io/, which can not only create 3d models from a single photo of a person, but also animate them! Also FaceBuilder for Blender can create detailed faces with just a couple photos: keentools.io/products/facebui...
February 22, 2022 NOTE: I got a few comments that the colab wasn't working. I just verified that it does still work. I tried the default image, then tried again with an image I uploaded of Keanu Reeves. The result is here: • Result from PIFuHD, Fe... Please make sure you follow the instructions exactly. If you get errors when uploading your own image, make sure that it does not exceed 60kb in size and that you go through each step after upload.
ANOTHER NOTE: Machine Learning progresses very quickly. The PIFuHD paper, code, and this colab are already quite "old". Newer tools now exist for transforming images (not just people) into 3d meshes, for example NVIDIA Omniverse/Kaolin. If you want to generate face meshes, StyleSDF (github.com/royorel/StyleSDF) has a colab that is very easy to use. I hope you will explore those other tools. Good luck!
Original Paper: shunsukesaito.github.io/PIFuHD/

Пікірлер: 110

@jackson501100 3 жыл бұрын

Ran into an issue w/ rendering. Was able to download the object, so that is exciting!!!! Thank you for this tutorial. It really helped. I went step by step pausing where needed. Thank you again sir! I appreciate you!

@Raksasa187 Жыл бұрын

amazing and super handy for background characters!! Thank you so much for the clear tutorial

@RiverReeves23 2 жыл бұрын

Thanks man, that is the most clear tutorial on how to do this. Someone else had a vid and it came up with errors. Your's worked perfectly!

@belkaosef7298 Жыл бұрын

getting to the point direcly ! thanks a lot man !

@jamesdutoit 2 жыл бұрын

Absolutely fantastic! Thank you, Sir.

@PambanSwamigalArul Жыл бұрын

Thanks a lot for the detailed tutorial

@user-vl8qg1vj8l 2 жыл бұрын

This is amazing!! Thank you!

@SuperMassman 3 жыл бұрын

Thank you,,, one more tool for my new hobby playing with Blender

@jareddinwiddie2332 2 жыл бұрын

So these can be used in blender?

@montymole297 2 жыл бұрын

@@jareddinwiddie2332 Yes, they can be downloaded as OBJ, which are files that Blender can import.

@vvortex157 Жыл бұрын

damn, that statue can become a creepy monster for a game

@Rain0015 3 жыл бұрын

Thanks

@SkyShazad Жыл бұрын

This is an Awesome Video my Dude.. My question is, When its done , and ive downloaded my model etc , and now i want to try the Process again do i need to start the whole process again??? as thats what im doing

@shadow_dancer 2 жыл бұрын

Interesting... One question - is it possible to use several pictures of the same person to get better mesh result at the end? Or this only works with single picture?

@brushrunner Жыл бұрын

anyone?

@panman69420 Жыл бұрын

no.

@wgalloPT 3 жыл бұрын

Question....how do i go by outputting multiple frames of a small animation? do i have to run it all over everytime?

@crider9260 2 жыл бұрын

Can you use multiple images to improve quality?

@moses5407 11 ай бұрын

Can circumferential and other measures be pulled from this directly ... Or is it via blender in some way?

@moesaadi4601 2 жыл бұрын

So I saw another video where the guy used the original photo and pretty much stuck it to the final result model to give it color and texture, would that work if the model is missing some of the areas?

@karubrandao Жыл бұрын

I'm trying to learn about ICON (implicit clothed humans obtained from normals) and Nvidia's nerf. Could you please try to do a tutorial (with texture) about these 2 because it seems they are the best till now for this purpose. Thank you!

@alizee3687 Жыл бұрын

Wao... fantastic I even run into one issue. Part of my image exploded in 3D transformation but everything was available in one frame i.e. leg was few cm away from main body and stomach was ripped open. I think it's because I choose low resolution image. Other then that all went fine

@daiesison Жыл бұрын

How do you run StyleSDF with your own images? On the collab it just runs with its own images.

@samchaw7837 2 жыл бұрын

i couldnt figure out , i am following all the instructions as you mentioned but my object files comes in sharttered pieces, not as one invidiual object, can you tell why its happening like that, thanks in advance

@clownworldhouse 3 жыл бұрын

Darn impressive

@justicemagagane5737 3 жыл бұрын

its ok bro you can say 'damn'

@clownworldhouse 3 жыл бұрын

@Justice Magagane d*mn 😯

@justicemagagane5737 3 жыл бұрын

@@clownworldhouse lol wtf

@T1mothyTee 2 жыл бұрын

Hello is there a way to process a bunch of images as a batch? I want to upload about 100 images of people at once and have it generate a mesh for each of them. Thanks!

@DanteS-119 Жыл бұрын

you need to run this on your own system and write a program to do that for you....

@badcapybara 2 жыл бұрын

Hey, great tut, thanks a lot for the effort! Does it also generate missing textures?

@montymole297 2 жыл бұрын

This tutorial shows how to add textures: kzfaq.info/get/bejne/oauCrct7sL_ad6M.html

@huytruong31127 Жыл бұрын

If i want create table or chair 3D, can I do with this model?

@koushikvallam9277 8 ай бұрын

I'm getting an error as "No distributions found for pytorch3d" after running pytorch3d line

@CyberPodMark2 11 ай бұрын

Will it work for 3d cartoon characters model from the picture?

@brushrunner Жыл бұрын

Is there any version of this for the textures?

@wendeng6279 Жыл бұрын

At the last step, I got error message: valueerror not enough values to unpack( expected 4, got 0), anybody have the same problem? thanks.

@johndavis1079 2 жыл бұрын

I'm doing it but it's not doing the image

@user-vl8qg1vj8l 2 жыл бұрын

I have one question! Is it possible with luggage?

@user-kz5pz2zi3q 5 ай бұрын

I need to know how to add movement to the 3d model .. i need it for sign language project

@MrNutmegdesigns Жыл бұрын

Is this only for people or can it be of object as well?

@C0smic_Star 2 жыл бұрын

Can this work with animated characters?

@butterflystudio606 2 жыл бұрын

Hello sir ! I faced some difficulties downloading this part net = PoseEstimationWithMobileNet() checkpoint = torch.load('checkpoint_iter_370000.pth', map_location='cpu') load_state(net, checkpoint) get_rect(net.cuda(), [image_path], 512) and I couldn't fin d any files in recon how can I FIX THAT

@exogendesign4582 3 ай бұрын

have you found fix? I also got this error. I wanted to turn a figure into 3d for trophy design.

@amattomy 3 жыл бұрын

here all the way from corridor crew

@jcbyy 3 жыл бұрын

haha same

@jonasloekke7696 2 жыл бұрын

Yep me too

@3dgiftz 2 жыл бұрын

😔

@DanFrederiksen 9 ай бұрын

it does make sense that a human shape is a fairly narrow expression space so it should actually be a fairly easy task. I would think it's vastly simpler than what midjourney does for instance. it should be quite possible to make 3D models with as good fidelity as our minds can. essentially perfect

@Padma226 Жыл бұрын

i am receiving NameError: name 'image_path' is not defined

@ramgladore Жыл бұрын

Kinda sorta worked for me but it still looked janky. Maybe the image I used too low rez.

@gustavodegrossi4708 3 жыл бұрын

Excelent video ! Do yo know any easy way to put the texture back of the person? Or I have to do it on cinema 4D? Thanks!!!!!

@TheAgentMeme 3 жыл бұрын

the tutorial to put the textures back on kzfaq.info/get/bejne/oauCrct7sL_ad6M.html&ab_channel=FreedomArts-3DAnimation%26GameDeveloper

@Justkev______ 3 жыл бұрын

that's the same thing i want to know

@tabz4040 2 жыл бұрын

Error while finding module specification for 'apps.simple_test' (ModuleNotFoundError: No module named 'apps')

@tabz4040 2 жыл бұрын

any fix?

@ruchalian 2 жыл бұрын

creepy AI can make in a second at the same time if this helpful in the future for medical it would be great overall what i think today was very interesting i wanna try this.

@pabloa.colchadop.7409 2 жыл бұрын

Hi. How can I unistall or remove the pytorch 3d??

@loscorralesdelasplazasdeto3586 2 жыл бұрын

For animls??

@PalKellyMedia 2 жыл бұрын

Please I need link to the website

@user-kz5pz2zi3q 5 ай бұрын

How to add movements to the 3d model

@amankhan6552 Жыл бұрын

I get this error when i run pifuhd ValueError: not enough values to unpack (expected 4, got 0)

@hiksms Жыл бұрын

same

@snippetshub 2 жыл бұрын

This isn't working, as the latest commenters also stated. As for me, the error is when running the "from google collab import files" cell. No upload button is enabled AFTER CLICKING THE START BUTTON, and I got this message: "Upload widget is only available when the cell has been executed in the current browser session. Please rerun this cell to enable." Seriously??

@ginnieware1633 Жыл бұрын

3 times I have tried this and 3 times failed apparently the link above gives me a figure that is already in the system and no instructions on how to clear the field.

@noodlesmarsden 6 ай бұрын

I'm slightly late to the party but what happens when I have no files in results but followed every step you did?

@sairajshinde5269 Жыл бұрын

it is not able to create obj file ,it is showing only the image .can someone help me with the issue plzz

@derekau533 Жыл бұрын

Just put in a fix in the colab, thanks for letting me know!

@dayonmohler5893 3 жыл бұрын

Can you do this if computer doesn’t have gpu?

@randomplayer6461 2 жыл бұрын

how do you use a pc if it doesn't have a gpu. (every pc has a gpu)

@DanFrederiksen 10 ай бұрын

doesn't it include textures?

@sajithavijay6686 3 ай бұрын

We can rige this ?

@chemonized Жыл бұрын

this doesn't work anymore, getting errors in your rendering section "could not find a version that satisfies the requirement" and on the Run PiFuHD section it gives this error: error cannot marching cubes cannot unpack non-iterable int object

@derekau533 Жыл бұрын

I have just updated the colab with fixes. Some libraries needed to be updated. Thanks for letting me know!

@chemonized Жыл бұрын

@@derekau533 man that's amazing so fast you updated it! Works great now! Thank you! Side question, it says maximum 60kb image, but can you run a higher resolution if you are making it use your own machine ?

@derekau533 Жыл бұрын

@@chemonized Yes, I think so.. bigger images just require more memory. You can try, then scale down if out of memory error occurs. Please remember this software is three years old now, which is ancient! The models are not super detailed. I'm sure there are better alternatives available now.

@seekerandroses Жыл бұрын

hey is this no longer working?

@user-ks9hb2zw3t 8 ай бұрын

not enough values to unpack

@tomasalmiron1487 Жыл бұрын

"recon" is empty. :(

@babubhaiyavines7077 10 ай бұрын

Same problem bro... Did you find solution???

@tiberiusgrozav1438 Жыл бұрын

Hi there. Cool video but i get following error when running PIFuHD: "error cannot marching cubes cannot unpack non-iterable int object" As output in results/recon i only have the .png file and not the .obj file. The input file was under 60Kb as you said. What am I doing wrong?

@DR-to5hb Жыл бұрын

I also am only getting the png in output. Errors began in the render cell, and the two suggested options in that cell do not work for me.

@derekau533 Жыл бұрын

I just updated the colab with a fix, thanks for letting me know!

@RuubRoyce 11 ай бұрын

You only have a front faced picture to work with, yet this "AI" protocol struggled to determine which way was front?!

@darkfantom5757 Жыл бұрын

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. torchvision 0.14.0+cu116 requires torch==1.13.0, but you have torch 1.6.0+cu101 which is incompatible. torchtext 0.14.0 requires torch==1.13.0, but you have torch 1.6.0+cu101 which is incompatible. torchaudio 0.13.0+cu116 requires torch==1.13.0, but you have torch 1.6.0+cu101 which is incompatible. fastai 2.7.10 requires torch=1.7, but you have torch 1.6.0+cu101 which is incompatible.

@darkfantom5757 Жыл бұрын

i do all by guide and model is broken

@velsein Жыл бұрын

doesn't work :/

@rajeevprabhu6722 2 жыл бұрын

You made a mistake by uploading a jpg pic. Process and upload a png or psd one with a trasperant background.

@JohnPaulBuce Жыл бұрын

ded link bro

@avantgaarden 6 ай бұрын

how does it work with non-human figures?

@derekau533 6 ай бұрын

this model is specifically for images of people. you can try another model like ZoeDepth huggingface.co/spaces/shariqfarooq/ZoeDepth

@avantgaarden 6 ай бұрын

thanks a ton this is helpful@@derekau533

@larrylurex86 Жыл бұрын

Tip: Instead of posting only how it works and how it looks the whole process it would be much more usefull to show what requirements are needed and how to set all the dependencies required by colab or to get the process working at least! If you don't do that it's just a random video about something that probably it works only for you and few other users but if you want to get much more attention you should add all the settings that make this process working! Cheers.

@johndavis1079 2 жыл бұрын

It doesn't work

@tronyxwall6378 2 жыл бұрын

its not working another useless thing

@derekau533 2 жыл бұрын

I just tested it today and can prove that the colab is working, please see the video description for updates.

@youtubedeletesmychannels2329 2 жыл бұрын

@@derekau533 it’s not working he’s right I got a error about a sample test idk wtf that means😂

@youtubedeletesmychannels2329 2 жыл бұрын

@@derekau533 eh idk care no more it’s too annoying how do I even reverse all the downloads and just uninstall the whole time waste all together

@vinayrana6632 2 жыл бұрын

@@derekau533 not working for me as well - facing this issue : ValueError: not enough values to unpack (expected 4, got 0)

@chrisw.3576 2 жыл бұрын

@@vinayrana6632 same here...

@Encumberedemotions 3 ай бұрын

This shit is slow and didn't work

@johanneszwilling 4 ай бұрын

Works now IF you fix the numpy issues. I had two with np.int and np.bool (/lib/sdf.py)

@destroyer7827 2 күн бұрын

what did u change exactly?

@jcoast1 Жыл бұрын

ERROR: Could not find a version that satisfies the requirement pytorch3d (from versions: none) ERROR: No matching distribution found for pytorch3d crap...dislike