3d-photo-inpainting
MiDaS
3d-photo-inpainting | MiDaS | |
---|---|---|
22 | 27 | |
6,828 | 4,089 | |
0.1% | 1.4% | |
0.0 | 2.4 | |
8 months ago | 3 months ago | |
Python | Python | |
GNU General Public License v3.0 or later | MIT License |
Stars - the number of stars that a project has on GitHub. Growth - month over month growth in stars.
Activity is a relative number indicating how actively a project is being developed. Recent commits have higher weight than older ones.
For example, an activity of 9.0 indicates that a project is amongst the top 10% of the most actively developed projects that we are tracking.
3d-photo-inpainting
- I have an AI Generated jpg. I want to add subtle looping animation to it
-
Whats the latest and greatest in 3d img2img/txt2img?
If you are looking to create actual 3d models, the DepthMap extension does have a function to create PLY models with vertex color information, and to render clips with simple camera moves from that extracted 3d scene, including inpainting (as per the 3d-photo-inpainting paper)
-
Quick test of AI and Blender with camera projection.
The depthmap extension for A1111 has implemented the 3d-photo-inpainting code that is doing that kind of thing. That's what I used to use, first on a Colab, and then adapted for windows so I could run it locally. But it's much more convenient to do it directly from the Automatic1111 WebUI.
- Is there an extension that does this?
-
Generate multiple complex subjects on a single image all at once with a depth aware custom extension!
But things are even older than stable diffusion.
-
Coronal mass ejection of the sun. Image from r/space. Crossview ML generated
It's a slightly modified version of https://shihmengli.github.io/3D-Photo-Inpainting/
-
[R] META researchers generate realistic renders from unseen views of any human captured from a single-view RGB-D camera
Thanks! I barely did anything though, just took a deep dream'ed photo made by another artist (Daniel Ambrosi) and passed it through this: https://shihmengli.github.io/3D-Photo-Inpainting/ (github and colab at bottom). Didn't even have to come up with the camera trajectory, was one of the presets in the repo
-
Tumultuous Seas
pretty sure it's this: https://github.com/vt-vl-lab/3d-photo-inpainting
- These are the raw frames I got from Gaugan2, but I'll be posting modified versions in the comment section.
- 3D Photography Using Context-Aware Layered Depth Inpainting
MiDaS
-
How to Estimate Depth from a Single Image
The checkpoint below uses MiDaS, which returns the inverse depth map, so we have to invert it back to get a comparable depth map.
-
Distance estimation from monocular vision using deep learning
Hi, I have made use of the KITTI dataset for this, and yes it depends on objects of know sizes. Here I have defined the following classes: Car, Van, Truck, Pedestrian, Person_sitting, Cyclist, Tram, Misc, or DontCare and the predictions are pretty accurate for those classes. Even if it's not the same class, it still recognizes the object since I have made use of the coco names dataset here and that is used along with YOLO for object detection. And there are several already implemented projects that make use of deep learning models trained on 2D datasets to predict 3D distance. This was one of my inspirations for this project: https://blogs.nvidia.com/blog/2019/06/19/drive-labs-distance-to-object-detection/ Furthermore, there are well-documented and researched papers like DistYOLO or MiDaS that makes use of deep learning for depth estimation
-
OMPR V0.6.10 update
-Added AI image depth generator Create your own depth map image at a click of a button. Using the awesome MIDAS3.1 https://github.com/isl-org/MiDaS as the backend and the model "dpt_beit_large_512" for the highest quality depth map. Video and GIF depth map generators coming out next together with the Depth movie player feature.
-
AI that converts a regular 2d image to stereoscopic
It uses MiDaS. That extension may be the most accessible way to use it at home. IDK.
-
Idea: training on magiceye images
Here's the project homepage https://github.com/isl-org/MiDaS
-
MiDaS v3_1 and DiscoDiffusion
The problem came up after MiDaS updated to version V3_1 on Dec 24th. Although the fix works fine, with the new version there are many changes, which for me produces slightly different results. I would like to able to produce results like before. I still clone the MiDaS repo, but then set it back to the last commit before the changes in december, which is 66882994a432727317267145dc3c2e47ec78c38a.
-
File not found error
try: from midas.dpt_depth import DPTDepthModel except: if not os.path.exists('MiDaS'): gitclone("https://github.com/isl-org/MiDaS.git") gitclone("https://github.com/bytedance/Next-ViT.git", f'{PROJECT_DIR}/externals/Next_ViT') if not os.path.exists('MiDaS/midas_utils.py'): shutil.move('MiDaS/utils.py', 'MiDaS/midas_utils.py') if not os.path.exists(f'{model_path}/dpt_large-midas-2f21e586.pt'): wget("https://github.com/intel-isl/DPT/releases/download/1_0/dpt_large-midas-2f21e586.pt", model_path) sys.path.append(f'{PROJECT_DIR}/MiDaS')
-
A quick demo to show how structurally coherent depth2img is compared to img2img using Automatic1111.
Cool. The repo for MiDaS is here. https://github.com/isl-org/MiDaS You can see that they partially trained the model on 3D movies Here's a list of the movies that were used to train it. I wonder if they'll be training a MiDaS v 4.0 as things have moved on quite a bit since it was released in Apr 2021?
-
Boosting Monocular Depth repo
We present a stand-alone implementation of our Merging Operator. This new repo allows using any pair of monocular depth estimations in our double estimation. This includes using separate networks for base and high-res estimations, using networks not supported by this repo (such as Midas-v3), or using manually edited depth maps for artistic use. This will also be useful for scientists developing CNN-based MDE as a way to quickly apply double estimation to their own network. For more details please take a look here.
-
DepthViewer is now live on Steam :)
I'll make the feature to export only the depthmap .png file. If you need the depthmap .png right now you can use the MiDaS python script.
What are some alternatives?
VQGAN-CLIP - Just playing with getting VQGAN+CLIP running locally, rather than having to use colab.
stable-diffusion-webui-depthmap-script - High Resolution Depth Maps for Stable Diffusion WebUI
cupscale - Image Upscaling GUI based on ESRGAN
DenseDepth - High Quality Monocular Depth Estimation via Transfer Learning
image-super-resolution - 🔎 Super-scale your images and run experiments with Residual Dense and Adversarial Networks.
stablediffusion - High-Resolution Image Synthesis with Latent Diffusion Models
Real-ESRGAN - Real-ESRGAN aims at developing Practical Algorithms for General Image/Video Restoration.
deeplearning4j-examples - Deeplearning4j Examples (DL4J, DL4J Spark, DataVec) [Moved to: https://github.com/deeplearning4j/deeplearning4j-examples]
caire - Content aware image resize library
DiverseDepth - The code and data of DiverseDepth
BoostingMonocularDepth
Insta-DM - Learning Monocular Depth in Dynamic Scenes via Instance-Aware Projection Consistency (AAAI 2021)