Pre-trained Models for Neural Inverse Knitting

We provide a few of our pre-trained models (as checkpoint directories):

• Different levels of real training data (with normal cross entropy):
  • 200 samples
  • 400 samples
  • 800 samples
  • All 1678 samples
• Base network variants:
  • Normal CE
  • MIL CE
  • Normal CE + Transfer Data (S+U Learning)
• Refiner Network with different mixing ratios:
  • α=0.1
  • α=1/3
  • α=0.5
  • α=2/3
  • α=0.9
• Complex Refiner Network (with skip connections):
  • α=1/3
  • α=0.5
  • α=2/3