fusheng-ji / miniproject

Mini-Project

Datasets

First download from Google Drive the dataset folder, and put it under the root path.

• Object-level

  • Objaverse-Xl
    • Description
    • Can be downloaded by the API provied in the Colab tutorial
  • MVImageNet
    • Download link
  • OminiObject3D
    • Description
    • Download link
    • Used cat: beauty_blender
  • Google Scanned Object
    • Descrption
    • Used: Weisshai_Great_White_Shark
  • NeRF-based dataset
    • Download link
    • LLFF
    • synthetic

• Scene-level

  • Scannet++
    • Download link
    • downloaded: DSLR only (2 MP): 112 GB
  • 3D-Future
    • Download link
    • Only contain Single view
    • Select randomly 6 scenes from train/test set
  • RealEstate10K
    • Download link
    • used: test/ec2745428712d42b

Visualization

Task: Visualize the multi-view camera viewpoints

This part using the Python + webasm/js/css/html and can be used instantly (does not need any C++ compilation).

pip install requirement.txt
python OminiObject3D.py / 3D_Future.py 
open your browser with localhost:8888

Epipolar Geometry

Task: Draw the epipolar line between two views

Using SIFT to find Keypoint, experiment on the book_002/008.png and book_002/010.png from dataset/OmniObject3D.

python epipolar_draw.py

Task: Do the stereo rectification for two views

python stereo_rectification.py

References

The main structure of Visualization Part is builded on Alex Yu's plenoctrees. Thanks for your wonderful work!