Test Brightness Augmentation

HACC2018 / ohia.ai

MIT License

4 stars 1 forks source link

Test Brightness Augmentation #31

Closed mattmotoki closed 5 years ago

mattmotoki commented 5 years ago

Determine the effectiveness of brightness augmentation.

Take photos of a single plant at multiple times of the day (different brightness levels)
Train a model with/without brightness augmentation
Plot performance vs. brightness level for both models.
Calculate average percent improvement

mattmotoki commented 5 years ago

Fine-tuning results show that mobilenetv1 performs the best.

           model   top_1_accuracy   top_3_accuracy   top_5_accuracy
mobilenetv1-1.00           0.6450           0.8559           0.9219 
mobilenetv2-1.00           0.5521           0.7786           0.8698 
mobilenetv2-1.30           0.5842           0.8290           0.8993 
mobilenetv2-1.40           0.6068           0.8490           0.9149 
    nasnetmobile           0.5590           0.8056           0.8854