Dataset Viewer
Auto-converted to Parquet Duplicate
run_name
stringclasses
6 values
accuracy
float64
0.46
0.64
precision
float64
0.38
0.75
recall
float64
0.06
0.85
f1_score
float64
0.1
0.67
roc_auc
float64
0.55
0.71
pr_auc
float64
0.54
0.74
epochs1_bs8_lr0p0001_183619
0.57
0.556962
0.846154
0.671756
0.564904
0.58451
epochs1_bs8_lr0p0001_185026
0.46
0.375
0.057692
0.1
0.553686
0.539457
epochs3_bs8_lr0p0001_190520
0.58
0.692308
0.346154
0.461538
0.566907
0.592806
epochs5_bs16_lr0p0001_190528
0.54
0.75
0.173077
0.28125
0.63742
0.691261
epochs8_bs16_lr5e-05_190540
0.64
0.75
0.461538
0.571429
0.707532
0.737011
epochs10_bs32_lr5e-05_192729
0.61
0.675676
0.480769
0.561798
0.659054
0.638121

Initial Evaluation Results

This repo contains the results from our Stage 1 baseline deepfake classifier

Runs included:

-# Initial Evaluation Results

Baseline ViT runs with different training settings:

  • epochs=1, batch_size=8, lr=1e-4
  • epochs=3, batch_size=8, lr=1e-4
  • epochs=5, batch_size=16, lr=1e-4
  • epochs=8, batch_size=16, lr=5e-5
  • epochs=10, batch_size=32, lr=5e-5

Best run:

  • epochs=8, batch_size=16, lr=5e-5
  • Accuracy: 0.64 | F1: 0.57 | ROC-AUC: 0.71

See all_runs_summary.csv for full comparison.

Downloads last month
16