performance_history.txt

1. basic architecture, no transformations, no class based weights, BCE: .85 accuracy, 0.22 F1

2. basic architecture, no transformations, class based weights, BCE: .85 accuracy, 0.28 F1

3. basic, normalization based on channel means and stdevs, class based weights, BCE: .85 accuracy, 0.26 F1

4. TINYVGG-based basic, normalization based on channel means and stdevs, class based weights, BCE: 0.87 accuracy, 0.37 F1

5. TINYVGG-based basic, normalization based on channel means and stdevs, class based weights, randomflips (hor and vert), BCE: 0.88 accuracy, 0.36 F1

6. TINYVGG-based basic, normalization based on channel means and stdevs, class based weights, randomflips (hor and vert), dropout layers, BCE: 0.88 accuracy, 0.30 F1

7. DENSENET PRETRAINED, normalization based on channel means and stdevs,dropout layers (already in premade), BCE: 0.90 accuracy, 0.59 F1