Skip to content

Latest commit

 

History

History
82 lines (69 loc) · 6.81 KB

LOGS.md

File metadata and controls

82 lines (69 loc) · 6.81 KB

Publications

  1. Rich Model for Steganalysis of Color Images
  2. The ALASKA Steganalysis Challenge: A First Step Towards Steganalysis ”Into The Wild”
  3. Rich Models for Steganalysis of Digital Images
  4. Pixels-off: Data-augmentation Complementary Solution for Deep-learning Steganalysis

Useful

  1. https://www.kaggle.com/remicogranne/jpeg-explanations
  2. https://github.com/digantamisra98/EvoNorm/blob/master/evonorm2d.py
  3. https://github.com/Steganalysis-CNN/residual-steganalysis/blob/master/init/res_finetune_test.m
  4. https://github.com/yedmed/steganalysis_with_CNN_Yedroudj-Net/blob/master/pytorch_version/Covariance_Pooling_Steganalytic_Network_cat.py

https://github.com/facebookresearch/detr/tree/master/models https://www.youtube.com/watch?v=v8U9mM1Vwv0 https://github.com/mahyarnajibi/FreeAdversarialTraining/blob/master/configs.yml https://github.com/TAMU-VITA/Adv-SS-Pretraining/tree/master/pretraining

Models

Experiment Name Model Fold bAUC cAUC Acc01 LB LB (Flip) LB (D4)
May07_16_48_rgb_resnet34_fold0 rgb_resnet34 0 8449 56.97
May07_16_48_rgb_resnet34_fold0 (fine-tune) rgb_resnet34 0 8451 56.90
May08_22_42_rgb_resnet34_fold1 rgb_resnet34 1 8439 56.62
May09_15_13_rgb_densenet121_fold0_fp16 rgb_densenet121 0 8658 8660 60.90
May11_08_49_rgb_densenet201_fold3_fp16 rgb_densenet201 3 8402 8405 56.38
---------------------------------------------- ------------------------ ------ ------ ------ ------- ----- ----------- ---------
May13_23_00_rgb_skresnext50_32x4d_fold0_fp16 rgb_skresnext50_32x4d 0 9032 9032 67.22
May13_19_06_rgb_skresnext50_32x4d_fold1_fp16 rgb_skresnext50_32x4d 1 9055 9055 67.60
May12_13_01_rgb_skresnext50_32x4d_fold2_fp16 rgb_skresnext50_32x4d 2 9049 9048 67.56
May11_09_46_rgb_skresnext50_32x4d_fold3_fp16 rgb_skresnext50_32x4d 3 8700 8699 61.45
---------------------------------------------- ------------------------ ------ ------ ------ ------- ------- ------- -------
May15_17_03_ela_skresnext50_32x4d_fold1_fp16 ela_skresnext50_32x4d 1 9144 9144 69.55 0.915 0.919 0.919
May21_13_28_ela_skresnext50_32x4d_fold2_fp16 ela_skresnext50_32x4d 2 9164 9163 70.17 0.921 0.921
May24_11_08_ela_skresnext50_32x4d_fold0_fp16 ela_skresnext50_32x4d 0
May26_12_58_ela_skresnext50_32x4d_fold3_fp16 ela_skresnext50_32x4d 3 0.922
------------------------------------------------ ------------------------- ------ ------ ------ ------- ------- ------- -------
May18_20_10_ycrcb_skresnext50_32x4d_fold0_fp16 ycrcb_skresnext50_32x4d 0 8266 8271 55.34
------------------------------------------------ ------------------------- ------ ------ ------ ------- ------- ------- -------
May28_13_04_rgb_tf_efficientnet_b6_ns_fold0 rgb_tf_efficientnet_b6 0 0.917
May28_18_49_rgb_tf_efficientnet_b6_ns_fold1 rgb_tf_efficientnet_b6 1 0.923

Models (New folds, Holdout)

rgb_tf_efficientnet_b6_ns

|----------------------------------------------|-----------------------------|------|--------|--------|--------|-------|-------|-----------|---------|

Experiment Name Model Fold Metric bAUC cAUC Acc01 LB LB (Flip) LB (D4)
Jun05_08_49_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 0 loss 0.9199 0.9200 71.36
Jun05_08_49_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 0 b-auc 0.9205 0.9205 70.93
Jun05_08_49_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 0 c-auc 0.9205 0.9205 70.93
---------------------------------------------- ----------------------------- ------ -------- -------- -------- ------- ------- ----------- ---------
Jun09_16_38_rgb_tf_efficientnet_b6_ns * rgb_tf_efficientnet_b6_ns 1 loss 0.9237 0.9238 72.29
Jun09_16_38_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 1 b-auc 0.9237 0.9238 72.29
Jun09_16_38_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 1 c-auc 0.9237 0.9238 72.29
---------------------------------------------- ----------------------------- ------ -------- -------- -------- ------- ------- ----------- ---------
Jun11_08_51_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 2 loss
Jun11_08_51_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 2 b-auc
Jun11_08_51_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 2 c-auc
---------------------------------------------- ----------------------------- ------ -------- -------- -------- ------- ------- ----------- ---------
Jun10_08_49_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 3 loss
Jun10_08_49_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 3 b-auc
Jun10_08_49_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 3 c-auc
---------------------------------------------- ----------------------------- ------ -------- -------- -------- ------- ------- ----------- ---------
Jun18_19_24_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 0 loss 0.9264 0.9254 72.33
Jun18_19_24_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 0 b-auc 0.9265 0.9253 72.08 0.926 / 0.924
Jun18_19_24_rgb_tf_efficientnet_b6_ns rgb_tf_efficientnet_b6_ns 0 c-auc 0.9264 0.9254 72.33 0.923 / 0.922
---------------------------------------------- ----------------------------- ------ -------- -------- -------- ------- ------- ----------- ---------

Average of 4 folds (best loss): Average of 4 folds (best auc b): Average of 4 folds (best auc c): Average of 4 folds (average of all 3):

https://github.com/YangzlTHU/IStego100K https://arxiv.org/pdf/1911.05542.pdf