Sekunde/Pri3D
[ICCV'21] Pri3D: Can 3D Priors Help 2D Representation Learning?
Pri3D helps computer vision engineers improve the performance of 2D image understanding tasks, such as identifying objects or segmenting images. By leveraging 3D depth and geometric information during an initial training phase, it creates more robust features from standard 2D images. The output is a pre-trained model that delivers better results for downstream 2D image analysis.
150 stars. No commits in the last 6 months.
Use this if you are developing computer vision models for tasks like semantic segmentation, object detection, or instance segmentation and want to improve their accuracy by incorporating 3D geometric insights.
Not ideal if your primary goal is 3D reconstruction, point cloud processing, or if you do not have access to RGB-D (color and depth) datasets for pre-training.
Stars
150
Forks
8
Language
Jupyter Notebook
License
—
Category
Last pushed
Dec 17, 2021
Commits (30d)
0
Get this data via API
curl "https://pt-edge.onrender.com/api/v1/quality/computer-vision/Sekunde/Pri3D"
Open to everyone — 100 requests/day, no key needed. Get a free key for 1,000/day.
Higher-rated alternatives
3DOM-FBK/deep-image-matching
Multiview matching with deep-learning and hand-crafted local features for COLMAP and other SfM...
suhangpro/mvcnn
Multi-view CNN (MVCNN) for shape recognition
zouchuhang/LayoutNet
Torch implementation of our CVPR 18 paper: "LayoutNet: Reconstructing the 3D Room Layout from a...
andyzeng/tsdf-fusion-python
Python code to fuse multiple RGB-D images into a TSDF voxel volume.
andyzeng/tsdf-fusion
Fuse multiple depth frames into a TSDF voxel volume.