Joy Chen, Qingfu Wan
This is the repository for the final project of "Machine Learning - CSCI-GA.2565-001" Spring 2021 at NYU Courant.
-
Conditioning the image generation on class labels for a better equilibrium in imbalanced multi-modal data.
-
Metrics-turned-Trainable-Scores + Self-aligned Perceptual Style Scores for better-learned data distribution.
-
Built-in classifier as an implicit evaluator and an explicit learner.
-
Label-aware class-specific generation + class-agnostic homogenous generation system.
-
A foolproof online self-annotation mechanism compensating for insufficient data.
Full report here.
${ROOT}
+-- experiments
+-- figs
+-- code
+-- full_code
+-- README.md
${ROOT}/code/
${ROOT}/full_code/
- PyTorch
- Keras
Left: Real Data; Right: Generated Data
Clear indication of ribs, spine, and heart shading.
Clear spines, lungs, and varied contrast & color gradient.