Recreational Image Reconstruction with Decision Trees
summary
Pipeline that maps pixel coordinates to RGB with per-channel decision trees, then replaces XY with distances to implanted sensors using Poisson-disk sensor placement so tree thresholds produce circular regions rather than rectangles. Trains on a 5% pixel subsample with numpy and scikit-learn, scales to random forests for smoother gradients via random forest aggregation, and weights samples by facial landmarks to preserve faces. Outputs were high-quality graphics used on wedding invitations, variability from sensor sampling requires saving intermediate images.
AI-written summary · about
chapters (10)
- 0:00Introduction and motivation
- 1:07Image reconstruction concepts
- 3:36Decision trees for image reconstruction
- 5:26First implementation and results
- 6:51Feature representation with sensors
- 8:39Model parameters exploration
- 10:26Training data weighting
- 11:11Project insights and takeaways
- 12:37Orthogonal elements and insights
- 13:21Conclusion and resources
related talks
31mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet -An Open Source Workflow
26mDetecting and classifying object images using ruby
24mTiling & Parallel Processing of Large Images
36mDeep Learning - Segmentation, Identification and Classification
46mML Use Cases at Think Big Analytics
37mUsing Exploratory Data Analysis to Discover Patterns
49mGraph Based Processing of Big Images
45mFrom Radar Echoes To Real-Time Recognition: Deep Learning on Another Planet an Open Source Workflow