Hybrid Vehicle Classification with Synthetic Data

PrototypeResearch

Augmented a low-data vehicle classifier with Stable Diffusion synthetic images; the hybrid model reached 88.3% accuracy, beating both real-only and synthetic-only baselines. With Wilfred Okajevo.

Classifier output: real and synthetic training mixSample data

van · 0.919100%

truck · 0.878700%

car · 0.969600%

bus · 0.727200%

car · 0.949400%

truck · 0.555500%

Tiles are schematic colour fields, not sample images. Confidence values are invented.

PrototypeAn interface sketch of this system, not a screenshot. Every figure, name and record in it is invented.

Built with: PyTorch, ResNet-18, Stable Diffusion

Source