👋 Hi, I’m @yccyenchicheng, a CS PhD student at UIUC.
👀 I’m interested in generative models.
License: MIT License
👋 Hi, I’m @yccyenchicheng, a CS PhD student at UIUC.
👀 I’m interested in generative models.
I d like to get the text-guided texturization demo please:)
Hi, in shape completion code I get a "No module named 'utils.qual_util'" error or the line "SDFusion = create_model(opt)". What could be the cause of this?
Could you provide the language-guided evaluation script for subsequent work metric alignment and referencing.
Very Thanks.
Thanks for your great. I would like to know when the code will be released?
Can you provide shapenet V1 dataset?
Hi @yccyenchicheng ,
Thank you for releasing the code for this very interesting work.
While trying to follow the README.md, I see that the file for preparing the data, ./launchers/launch_create_sdf_shapenet.sh
is unavailable.
Could you please upload this?
Thanks
Abhinanda
Hi,
While trying to run demo_img2shape.ipynb
I encountered the following error, in the cell nr 3 (creating the model):
FileNotFoundError: [Errno 2] No such file or directory: 'random_path/SDFusion/external/clip/bpe_simple_vocab_16e6.txt.gz'
Any help with this one?
Thanks!
HI @yccyenchicheng ,
I didn't find the file "launch_create_sdf_snet.sh". Could you please to share it?
Bests,
Yingjie
Thank you for your well organized code. How can I train this code by multiple GPUs? Looking forward your reply, thiank you!
Hi, can you help me solve it? thank you!
Hello, thank you for sharing your great work. I would like to know which categories you trained on shapenet when doing multi-modal generation. How much GPU memory does training take up? How long did it take to train?
Hi,
Thank you for this great work! I have a question about the training process of text-to-shape. According to your paper, the model is trained to generate only the geometry of objects, so how do you handle the color/texture-related texts in the text-to-shape dataset? Did you just let the conditioned diffusion model learns to ignore such information and just capture the geometry-related information? Or did you apply other preprocess to eliminate color-related words in a given text?
Hello @yccyenchicheng,
I apologize for the inconvenience and hope this message finds you well. I find your research to be highly intriguing and I have some inquiries regarding the experimental details.
Specifically, I would like to inquire about the device(s) used when training the Sdfusion-mm2shape
. Additionally, I am curious to know how long it took to complete the training process.
Furthermore, I would like to kindly request if you could consider releasing the training code for Sdfusion-mm2shape
as soon as possible.
Thank you very much for your time and consideration. I appreciate any assistance you can provide!
Best regards
Hi,
Impressive work. Is it possible to give the dataset collection and the preprocessing step that went into SDF creation. Because most of the buildings in BuildingNet are open and have holes in them. When i tried SDF with them for a different project of mine, it did not work well. If you already have a subset of buildings that you have collected and preprocessed to make it a closed surface would like to use it for one of my projects.
./launchers/launch_create_sdf_pix3d.sh: No such file or directory
Hello!
There is no code to process pix3d dataset (need it to train for mesh generation).
Hi,
I am trying to install dependency packages for SDFusion, but getting an error in the pytorch3D installation.
I am using cuda 11.7 and nvidia driver 515.
What environment did you use for inference? like cuda, nvidia-driver version etc.
Hi,
Appreciate your great work! I notice that you compare the text-guided generation results with AutoSDF. Can you provide your checkpoint of AutoSDF on the text2shape dataset? It seems that AutoSDF only provide checkpoint on a different text-guided dataset (ShapeGlot?). Thank you!
Great work and thanks for sharing! I am trying to increase the resolution in vqvae_snet.yaml, but it seems that setting the resolution more than 64 quickly explode the memory and caused out of video memory. Is there any suggestion to increase the resolution for higher quality models? thanks.
hellow, how to solve the question,please
When i try to train train_sdfusion_snet,I met this problem.I have trained the vqave model on the ShapeNet model successfully.How can i fix it?
Hi, I ran into this issue when trying out the notebook for unconditional generation and shape completion. It seems that there is no "sdf_to_pcd" function in util_3d.py.
I want to use your model to train my dataset.Can you tell me how to transfer my dataset to the format in the model,liking the format in the ShapeNet_filelists.Thank you
Why do not use clip image and text encoder? In Stable diffusion, text encoder used clip text encoder, and freeze condtion model while training diffusion. But in SDFusion, text encoder do not use clip-text-encoder and text-encoder params do not freeze.
Could you explain it ? Thanks.
Hi, hope you can help in solving this error:
when running the third step in training for single-view reconstruction:
./launchers/train_sdfusion_img2shape.sh
I get this error:
Traceback (most recent call last):
File "train.py", line 172, in
train_main_worker(opt, model, train_dl, test_dl, test_dl_for_eval, visualizer, device)
File "train.py", line 51, in train_main_worker
model.set_input(data)
File "/home/paperspace/sdf/SDFusion/models/sdfusion_img2shape_model.py", line 240, in set_input
self.img = input['img']
KeyError: 'img'
The conclusion I've come to so far is that the code is written for training on pix3d (in step 3 of training) while I'm training on ShapeNet v1. The dictionary in the file snet_dataset.py does not contain a key called 'img' like the one in pix3d_dataset.py but the error shows it's still expected as an argument.
I tried adding the img in the dictionary but it didn't work.The previous 2 steps in training work fine, but I'm having issues here. Thanks in advance...
Why text condition model (bert) do not use pretrained CLIP model and freeze it while training?
In Stable diffusion, text condition model used pretrained CLIP/L-14 and freeze it while training.
Could you help explain it? Thanks.
in preprocess
dir, there is no file ./launchers/launch_create_sdf_pix3d.sh
, can you provide the script? thanks so much!
in models/sdfusion_mm_model.py
File "/root/h/SDFusion/models/sdfusion_mm_model.py", line 25, in <module> from models.networks.resnet_v1 import resnet18 ModuleNotFoundError: No module named 'models.networks.resnet_v1'
can you provide the related codes? thanks so much!
Hi @yccyenchicheng , I am interested in using the code "Text Guided Texturization/Completion" for my research work on generative 3D. I would like to kindly request for your assistance in sharing the code please?
hi, can you help me solve it?
The OSS license file for Threading Building Blocks 2018 Update 6 is currently missing.
https://github.com/yccyenchicheng/SDFusion/tree/master/preprocess/isosurface/tbb/tbb-2018_U6
Appreciate your great work!
When I try to run your text2shape model, I notice that stride=(1, 2, 2) in Downsample() and Upsample().
Could you tell me more reasons why you chose stride=(1, 2, 2)? Why not stride=(2, 2, 2)?
It seems that there is no evaluation code available. The indicators in the paper cannot be tested for alignment.
Could you provide evaluation code?
Very thanks.
FileNotFoundError: [Errno 2] No such file or directory: '../data/ShapeNet/SDF_v1/resolution_64/02691156/1a04e3eab45ca15dd86060f189eb133/isosurf.sdf'
Hi @yccyenchicheng ,
I have some question about multi-modalities. Which dataset is this portion of the function implemented on? and When the code for this part will be release ?
Best regards,
Yingjie CAI
I find no match file with launch_create_sdf_pix3d.sh. Could you please re-upload this file?
Thank you very much
Can you provide the file of the pre trained model? Wget prompt "Connecting to uofi. app. box.com (uofi. app. box.com) | 59.24.3.173 |: 443... failed: Unknown error."
Very good job!
The script launch_create_sdf_pix3d.sh is missing, how to get it?
The folder pix3d/filelists is missing, how to get it?
Looking forward to your reply!
Which is better between Triplane representation(Get3D) and Voxel Grid representation(SDFusion)?
What are Advantages and disadvantages of different 3D representation methods?
Thanks.
Thank you for your work and for open-sourcing the code! I am trying to run the demo_txt2shape notebook, but it seems like it's missing the model for vqvae-snet-all.pth
. Did I miss something? Thanks for the help!
What your work need which GPU and how much time for training?
Dear Yen-Chi,
thanks for this great work and open-sourcing the code and model checkpoints!
As there are no official data splits for Pix3D, except for the chair category, could you provide the data splits which were used in the paper?
Additionally, would it be possible for you to provide the per-class results of the single-view reconstruction results (Tab. 2) or even an evaluation script for that part?
Thanks a lot!
The result in this work is amazing and I'm very interested in the building reconstruction.
I try to train the vqvae and sdfusion models on the BuildingNet dataset, but the related scripts or configs needed by the buildingnet dataset is not provided.
Could you please provide the training files related to the buildingnet dataset?
Hi, @yccyenchicheng.
Appreciate your great work! I want to get more geometric details of the generated ShapeNet models. One direct way is to increase the resolution of the SDF to 128, but this requires more memory and computational resources. To reduce memory consumption, it may be necessary to reduce the dimensionality of the latent space. However, I am not sure if reducing the dimensionality of the latent space can result in a good VQVAE model? Do you have any suggestions on this? Or are there other ways to get more geometric details of the generated model?
A declarative, efficient, and flexible JavaScript library for building user interfaces.
🖖 Vue.js is a progressive, incrementally-adoptable JavaScript framework for building UI on the web.
TypeScript is a superset of JavaScript that compiles to clean JavaScript output.
An Open Source Machine Learning Framework for Everyone
The Web framework for perfectionists with deadlines.
A PHP framework for web artisans
Bring data to life with SVG, Canvas and HTML. 📊📈🎉
JavaScript (JS) is a lightweight interpreted programming language with first-class functions.
Some thing interesting about web. New door for the world.
A server is a program made to process requests and deliver data to clients.
Machine learning is a way of modeling and interpreting data that allows a piece of software to respond intelligently.
Some thing interesting about visualization, use data art
Some thing interesting about game, make everyone happy.
We are working to build community through open source technology. NB: members must have two-factor auth.
Open source projects and samples from Microsoft.
Google ❤️ Open Source for everyone.
Alibaba Open Source for everyone
Data-Driven Documents codes.
China tencent open source team.