AI_ComputerVision

• Inspired by Dr Kai Han
• For recording the progress of CV learning

Dataset:

Expand

coco

1. https://cocodataset.org/#download
2. https://github.com/cocodataset/cocoapi.git

Data Preprocessing

1. Tiny image strategy:

i) https://groups.csail.mit.edu/vision/TinyImages/
ii) https://openreview.net/pdf?id=s-e2zaAlG3I

Traditional Methods

Expand

Stitching

  SIFT and Harris corner detection
  https://www.cs.tau.ac.il/~turkel/imagepapers/comparison_sift-harris-corner.pdf

Detection:

  Line Detection:

  Hough Transform (HT)
  APAI3010

  Progressive Probabilistic Hough Transform (PPHT):
  It is an improvement over the traditional Hough Transform and works faster because it examines a randomly chosen subset of points with every iteration.

  Randomized Hough Transform (RHT):
  This algorithm randomly selects points from an image and constructs line segments, and therefore can reduce computation time while still maintaining accuracy.

  Radon Transform:
  This transform is designed to detect straight lines within an image and can be used on binary images to increase the robustness of Hough transform. It transforms an image into a parameter space where the presence of a line is easier to detect.

  Line Segment Detector (LSD):
  It is an edge-based line detection algorithm that can detect multiple straight lines within an image in real-time.

Text to img

Expand

1. Stable Diffusion 2. Lora / Dreambooth: low rank finetuning methods

Text to 3D

Expand

Diffusion models

1. DreamFusion:

   Nerf + Stable Diffusion + DMTet 1 hour for 1 case

   https://dreamfusion3d.github.io They didn't provide the official code, but there is a reliable third-party reproduction you can leverage: https://github.com/ashawkey/stable-dreamfusion

   Personal Ammendment https://github.com/Justinfungi/stable-dreamfusion/tree/HKUproject