Skip to content

Latest commit

 

History

History
15 lines (9 loc) · 962 Bytes

short_report.md

File metadata and controls

15 lines (9 loc) · 962 Bytes

Classify Fake or Real Images

##Approach A In file "classifier_conv32_drop0.5.png" is shown the model attached at the end of the Encoder model and then inside "train_classifier_log_classifier1.csv" there are the metrics/statistics of the learning process. The metrics that are used are AUC, Precision, Recall, accuracy, loss for validation and train sets.

##Approach B Check file "classifier_conv64_drop.2.png" where you can see the architecture of the model stacked with the Encoder and inside "train_classifier_log_classifier2.csv" you can observe the whole learning/training process for 70 epochs and batch size =128. Still the same metrics are been used.

##Results: From the two approaches the B (second) one performs better, lookin at the last line inside "train_classifier_log_classifier2.csv" -> trainAUC =0.99,trainAcc=0.99,trainPrecision=0.99,trainRecall=0.99.

Notice:

both approaches using the same batchSize and number of Epochs.