Synthesizing Filamentary Structured Images with GANs

Empirical Experiments

Quantitative result

Table 1: Quantitative results of segmentation performance of a baseline segmentation method on the same DRIVE (or STARE, HRF, NeuB1) test set while training on a different training set. Results are evaluated by average F1-score (%).

evaluation_table

Visual Results

  • Exemplar DRIVE phantoms generated by Fila-GAN

  • Fila-GAN drive visual result
  • Exemplar STARE phantoms generated by Fila-GAN

  • Fila-GAN stare visual result
  • Exemplar HRF phantoms generated by Fila-GAN

  • Fila-GAN hrf visual result
  • Exemplar 2D Neuron phantoms generated by Fila-GAN

  • Fila-GAN 2d neuron visual result
  • Exemplar DRIVE phantoms generated by Fila-sGAN

  • Fila-sGAN visual result