deepguard 0.2.4__tar.gz → 0.3.0__tar.gz

This diff represents the content of publicly available package versions that have been released to one of the supported registries. The information contained in this diff is provided for informational purposes only and reflects changes between package versions as they appear in their respective public registries.
Files changed (86) hide show
  1. {deepguard-0.2.4 → deepguard-0.3.0}/PKG-INFO +28 -28
  2. {deepguard-0.2.4 → deepguard-0.3.0}/README.md +28 -28
  3. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard.egg-info/PKG-INFO +28 -28
  4. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/explainer/base_explainer.py +4 -1
  5. {deepguard-0.2.4 → deepguard-0.3.0}/pyproject.toml +1 -1
  6. {deepguard-0.2.4 → deepguard-0.3.0}/LICENSE +0 -0
  7. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/__init__.py +0 -0
  8. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/data/__init__.py +0 -0
  9. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/data/dataset.py +0 -0
  10. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/data/dataset_cutmix.py +0 -0
  11. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/data/dataset_cutout.py +0 -0
  12. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/data/dataset_mixup.py +0 -0
  13. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/data/handle_imbalance.py +0 -0
  14. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/data/split_data.py +0 -0
  15. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/data/transforms.py +0 -0
  16. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/__init__.py +0 -0
  17. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/attention.py +0 -0
  18. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/drop.py +0 -0
  19. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/featextractor.py +0 -0
  20. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/global_query.py +0 -0
  21. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/mlp.py +0 -0
  22. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/patch.py +0 -0
  23. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/reducesize.py +0 -0
  24. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/se.py +0 -0
  25. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/transformer.py +0 -0
  26. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/weight_init.py +0 -0
  27. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/layers/window.py +0 -0
  28. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/models/__init__.py +0 -0
  29. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/models/ms_eff_gcvit.py +0 -0
  30. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/models/ms_eff_vit.py +0 -0
  31. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/optimizer/__init__.py +0 -0
  32. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/optimizer/build_optimizer.py +0 -0
  33. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/scheduler/__init__.py +0 -0
  34. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/scheduler/fetch_scheduler.py +0 -0
  35. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/utils/__init__.py +0 -0
  36. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/utils/build_metrics.py +0 -0
  37. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/utils/early_stopping.py +0 -0
  38. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/utils/meter.py +0 -0
  39. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard/utils/seed.py +0 -0
  40. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard.egg-info/SOURCES.txt +0 -0
  41. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard.egg-info/dependency_links.txt +0 -0
  42. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard.egg-info/requires.txt +0 -0
  43. {deepguard-0.2.4 → deepguard-0.3.0}/deepguard.egg-info/top_level.txt +0 -0
  44. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/__init__.py +0 -0
  45. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/cam_evaluator.py +0 -0
  46. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/explainer/cam_explainer.py +0 -0
  47. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/explainer/eigenvalue_explainer.py +0 -0
  48. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/explainer/gradient_explainer.py +0 -0
  49. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/explainer/misc_explainer.py +0 -0
  50. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/explainer/perturbation_explainer.py +0 -0
  51. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/metrics/__init__.py +0 -0
  52. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/metrics/cam_mult_image.py +0 -0
  53. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/metrics/perturbation_confidence.py +0 -0
  54. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/metrics/road.py +0 -0
  55. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/utils/image.py +0 -0
  56. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/utils/model_targets.py +0 -0
  57. {deepguard-0.2.4 → deepguard-0.3.0}/explainability/utils/reshape_transforms.py +0 -0
  58. {deepguard-0.2.4 → deepguard-0.3.0}/inference/__init__.py +0 -0
  59. {deepguard-0.2.4 → deepguard-0.3.0}/inference/image_predictor.py +0 -0
  60. {deepguard-0.2.4 → deepguard-0.3.0}/inference/image_predictor_prt.py +0 -0
  61. {deepguard-0.2.4 → deepguard-0.3.0}/inference/predict_video.py +0 -0
  62. {deepguard-0.2.4 → deepguard-0.3.0}/inference/utils.py +0 -0
  63. {deepguard-0.2.4 → deepguard-0.3.0}/inference/video_predictor.py +0 -0
  64. {deepguard-0.2.4 → deepguard-0.3.0}/inference/video_predictor_prt.py +0 -0
  65. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/__init__.py +0 -0
  66. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/celeb_df_v2/__init__.py +0 -0
  67. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/celeb_df_v2/crop_face.py +0 -0
  68. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/celeb_df_v2/detect_original_face.py +0 -0
  69. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/celeb_df_v2/split_data.py +0 -0
  70. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/celeb_df_v2/utils.py +0 -0
  71. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/face_detector.py +0 -0
  72. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/ff++/__init__.py +0 -0
  73. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/ff++/crop_face.py +0 -0
  74. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/ff++/detect_original_face.py +0 -0
  75. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/ff++/split_data.py +0 -0
  76. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/ff++/utils.py +0 -0
  77. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/frame_extractor.py +0 -0
  78. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/kodf/__init__.py +0 -0
  79. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/kodf/crop_face.py +0 -0
  80. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/kodf/detect_original_face.py +0 -0
  81. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/kodf/generate_metadata.py +0 -0
  82. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/kodf/split_data.py +0 -0
  83. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/kodf/utils.py +0 -0
  84. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/landmark_detector.py +0 -0
  85. {deepguard-0.2.4 → deepguard-0.3.0}/preprocess/yolov8n-face.pt +0 -0
  86. {deepguard-0.2.4 → deepguard-0.3.0}/setup.cfg +0 -0
{deepguard-0.2.4 → deepguard-0.3.0}/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: deepguard
3
- Version: 0.2.4
3
+ Version: 0.3.0
4
4
  Summary: Multi-Scale Efficient Global Context Vision Transformer for Robust Deepfake Detection
5
5
  Author-email: seoyunje <seoyunje2001@gmail.com>
6
6
  License: MIT
@@ -55,8 +55,8 @@ Dynamic: license-file
55
55
  </p>
56
56
 
57
57
  <p align="center">
58
- <a href="README_KR.md"><b>🇰🇷 한국어 버전</b></a> |
59
- <a href="README_JP.md"><b>🇯🇵 日本語版</b></a> |
58
+ <a href="https://github.com/HanMoonSub/DeepGuard/blob/main/README_KR.md"><b>🇰🇷 한국어 버전</b></a> |
59
+ <a href="https://github.com/HanMoonSub/DeepGuard/blob/main/README_JP.md"><b>🇯🇵 日本語版</b></a> |
60
60
  <a href="#-model-evaluation"><b>📈 Model Evaluation</b></a> |
61
61
  <a href="#-predict-image--video"><b>🔮 Try Demo</b></a>
62
62
  </p>
@@ -102,9 +102,9 @@ To evaluate the generalization and robustness of our deepfake detection model, w
102
102
 
103
103
  | Dataset | Real Videos | Fake Videos | Year | Participants | Description (Paper Title) | Details |
104
104
  | :--- | :---: | :---: | :---: | :---: | :--- | :---: |
105
- | **Celeb-DF-v2** | 890 | 5,639 | 2019 | 59 | *A Large-scale Challenging Dataset for DeepFake Forensics* | [🔗 Readme](preprocess/celeb_df_v2/README.md) |
106
- | **FaceForensics++** | 1,000 | 6,000 | 2019 | 1,000 | *Learning to Detect Manipulated Facial Images* | [🔗 Readme](preprocess/ff++/README.md) |
107
- | **KoDF** | 62,166 | 175,776 | 2020 | 400 | *Large-Scale Korean Deepfake Detection Dataset* | [🔗 Readme](preprocess/kodf/README.md) |
105
+ | **Celeb-DF-v2** | 890 | 5,639 | 2019 | 59 | *A Large-scale Challenging Dataset for DeepFake Forensics* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/celeb_df_v2/README.md) |
106
+ | **FaceForensics++** | 1,000 | 6,000 | 2019 | 1,000 | *Learning to Detect Manipulated Facial Images* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/ff++/README.md) |
107
+ | **KoDF** | 62,166 | 175,776 | 2020 | 400 | *Large-Scale Korean Deepfake Detection Dataset* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/kodf/README.md) |
108
108
 
109
109
  <div id="data-preparation"></div>
110
110
 
@@ -154,9 +154,9 @@ DATA_ROOT/
154
154
 
155
155
  Click the links below to view the specific preprocessing details for each dataset:
156
156
 
157
- * [**Celeb-DF V2 Preprocess**](/preprocess/celeb_df_v2/README.md)
158
- * [**FaceForensics++ Preprocess**](/preprocess/ff++/README.md)
159
- * [**KoDF Preprocess**](/preprocess/kodf/README.md)
157
+ * [**Celeb-DF V2 Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/celeb_df_v2/README.md)
158
+ * [**FaceForensics++ Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/ff++/README.md)
159
+ * [**KoDF Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/kodf/README.md)
160
160
 
161
161
  ## 🏗 Model Architecture
162
162
 
@@ -164,9 +164,9 @@ Multi Scale Efficient Global Context Vision Transformer is an optimized multi-sc
164
164
 
165
165
  #### Explore More Details
166
166
 
167
- - [Model Architecture: MS-EffViT](deepguard/MS_EffViT.md) - _**Multi Scale Efficient Vision Transformer**_
167
+ - [Model Architecture: MS-EffViT](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/MS_EffViT.md) - _**Multi Scale Efficient Vision Transformer**_
168
168
 
169
- - [Advanced Architecture: MS-EFFGCViT](deepguard/MS_EffGCViT.md) - _**Multi Scale Efficient Global Context Vision Transformer**_
169
+ - [Advanced Architecture: MS-EFFGCViT](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/MS_EffGCViT.md) - _**Multi Scale Efficient Global Context Vision Transformer**_
170
170
 
171
171
 
172
172
  <p align="center">
@@ -181,15 +181,15 @@ We utilizes two distinct types of self-attention to capture both long-range and
181
181
 
182
182
 
183
183
  <p align="center">
184
- <img src=https://raw.githubusercontent.com/HanMoonSub/DeepGuard/main/docs/architectures/window_attention.JPG width="100%" height="300">
184
+ <img src=docs/architectures/window_attention.JPG width="100%" height="300">
185
185
  </p>
186
186
 
187
187
  ## 🧬 Model Zoo
188
188
 
189
189
  | Model | Resolution | # Total Params(M) | # Backbone(M) | # L-ViT(M) | # H-ViT(M) | FLOPs (G) | Model Config |
190
190
  | ----- | ---------- | -------------- | ----------- |------------- | ------------- | -------------- | ------- |
191
- | ⚡ ms_eff_gcvit_b0 | 224 X 224 | 8.7 | 3.6(41.4%) | 1.7(19.5%) | 3.3(37.9%) | 0.87 | [spec](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
192
- | 🔥 ms_eff_gcvit_b5 | 384 X 384 | 50.3 | 27.3(54.3%) | 6.6(13.1%) | 16.1(32.0%) | 13.64 | [spec](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
191
+ | ⚡ ms_eff_gcvit_b0 | 224 X 224 | 8.7 | 3.6(41.4%) | 1.7(19.5%) | 3.3(37.9%) | 0.87 | [spec](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
192
+ | 🔥 ms_eff_gcvit_b5 | 384 X 384 | 50.3 | 27.3(54.3%) | 6.6(13.1%) | 16.1(32.0%) | 13.64 | [spec](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
193
193
 
194
194
  ## 🚀 Training
195
195
 
@@ -231,22 +231,22 @@ We provide training scripts for both `ms_eff_vit` and `ms_eff_gcvit`. We recomme
231
231
 
232
232
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
233
233
  | ------------- | -------- | -------- | ---------- | -------- | ----- |
234
- | ms_eff_gcvit_b0 | 0.9842 | 0.9965 | 0.0283 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_celeb_df_v2.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
235
- | ms_eff_gcvit_b5 | 0.9981 | 0.9984 | 0.0089 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_celeb_df_v2.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
234
+ | ms_eff_gcvit_b0 | 0.9842 | 0.9965 | 0.0283 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_celeb_df_v2.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
235
+ | ms_eff_gcvit_b5 | 0.9981 | 0.9984 | 0.0089 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_celeb_df_v2.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
236
236
 
237
237
  **FaceForensics++ Pretrained Models**
238
238
 
239
239
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
240
240
  | ------------- | -------- | -------- | ---------- | -------- | ------ |
241
- | ms_eff_gcvit_b0 | 0.9808 | 0.9969 | 0.0637| [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_ff++.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
242
- | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_ff++.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
241
+ | ms_eff_gcvit_b0 | 0.9808 | 0.9969 | 0.0637| [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_ff++.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
242
+ | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_ff++.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
243
243
 
244
244
  **KoDF Pretrained Models**
245
245
 
246
246
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
247
247
  | ------------- | -------- | -------- | ---------- | -------- | ------ |
248
- | ms_eff_gcvit_b0 | 0.9655 | 0.9792 | 0.1237 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b0_kodf.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
249
- | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b5_kodf.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
248
+ | ms_eff_gcvit_b0 | 0.9655 | 0.9792 | 0.1237 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b0_kodf.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
249
+ | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b5_kodf.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
250
250
 
251
251
  ## 💻 Model Usage
252
252
 
@@ -499,14 +499,14 @@ result = explainer.display_heatmap_bbox_on_image(
499
499
 
500
500
  The jupyter notebooks themselves can be found under the tutorials folder in the git repository.
501
501
 
502
- - [Notebook tutorial: Predict DeepFake Image with ImagePredictor](./tutorials/predict_image.ipynb)
503
- - [Notebook tutorial: Predict DeepFake Video with VideoPredictor](./tutorials/predict_video.ipynb)
504
- - [Notebook tutorial: Low-Level Branch XAI Visualization (HiResCAM, GradCAMElementWise, LayerCAM)](./tutorials/low_level_visualization.ipynb)
505
- - [Notebook tutorial: High-Level Branch XAI Visualization (EigenGradCAM, GradCAM++, XGradCAM)](./tutorials/high_level_visualization.ipynb)
506
- - [Notebook tutorial: MS-EffViT Low-Level Branch Explainability](./tutorials/ms_eff_vit_low_level.ipynb)
507
- - [Notebook tutorial: MS-EffViT High-Level Branch Explainability](./tutorials/ms_eff_vit_high_level.ipynb)
508
- - [Notebook tutorial: MS-EffGCViT Low-Level Branch Explainability](./tutorials/ms_eff_gcvit_low_level.ipynb)
509
- - [Notebook tutorial: MS-EffGCViT High-Level Branch Explainability](./tutorials/ms_eff_gcvit_high_level.ipynb)
502
+ - [Notebook tutorial: Predict DeepFake Image with ImagePredictor](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/predict_image.ipynb)
503
+ - [Notebook tutorial: Predict DeepFake Video with VideoPredictor](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/predict_video.ipynb)
504
+ - [Notebook tutorial: Low-Level Branch XAI Visualization (HiResCAM, GradCAMElementWise, LayerCAM)](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/low_level_visualization.ipynb)
505
+ - [Notebook tutorial: High-Level Branch XAI Visualization (EigenGradCAM, GradCAM++, XGradCAM)](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/high_level_visualization.ipynb)
506
+ - [Notebook tutorial: MS-EffViT Low-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_vit_low_level.ipynb)
507
+ - [Notebook tutorial: MS-EffViT High-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_vit_high_level.ipynb)
508
+ - [Notebook tutorial: MS-EffGCViT Low-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_gcvit_low_level.ipynb)
509
+ - [Notebook tutorial: MS-EffGCViT High-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_gcvit_high_level.ipynb)
510
510
 
511
511
  ## 📬 Authors
512
512
 
{deepguard-0.2.4 → deepguard-0.3.0}/README.md RENAMED
@@ -29,8 +29,8 @@
29
29
  </p>
30
30
 
31
31
  <p align="center">
32
- <a href="README_KR.md"><b>🇰🇷 한국어 버전</b></a> |
33
- <a href="README_JP.md"><b>🇯🇵 日本語版</b></a> |
32
+ <a href="https://github.com/HanMoonSub/DeepGuard/blob/main/README_KR.md"><b>🇰🇷 한국어 버전</b></a> |
33
+ <a href="https://github.com/HanMoonSub/DeepGuard/blob/main/README_JP.md"><b>🇯🇵 日本語版</b></a> |
34
34
  <a href="#-model-evaluation"><b>📈 Model Evaluation</b></a> |
35
35
  <a href="#-predict-image--video"><b>🔮 Try Demo</b></a>
36
36
  </p>
@@ -76,9 +76,9 @@ To evaluate the generalization and robustness of our deepfake detection model, w
76
76
 
77
77
  | Dataset | Real Videos | Fake Videos | Year | Participants | Description (Paper Title) | Details |
78
78
  | :--- | :---: | :---: | :---: | :---: | :--- | :---: |
79
- | **Celeb-DF-v2** | 890 | 5,639 | 2019 | 59 | *A Large-scale Challenging Dataset for DeepFake Forensics* | [🔗 Readme](preprocess/celeb_df_v2/README.md) |
80
- | **FaceForensics++** | 1,000 | 6,000 | 2019 | 1,000 | *Learning to Detect Manipulated Facial Images* | [🔗 Readme](preprocess/ff++/README.md) |
81
- | **KoDF** | 62,166 | 175,776 | 2020 | 400 | *Large-Scale Korean Deepfake Detection Dataset* | [🔗 Readme](preprocess/kodf/README.md) |
79
+ | **Celeb-DF-v2** | 890 | 5,639 | 2019 | 59 | *A Large-scale Challenging Dataset for DeepFake Forensics* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/celeb_df_v2/README.md) |
80
+ | **FaceForensics++** | 1,000 | 6,000 | 2019 | 1,000 | *Learning to Detect Manipulated Facial Images* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/ff++/README.md) |
81
+ | **KoDF** | 62,166 | 175,776 | 2020 | 400 | *Large-Scale Korean Deepfake Detection Dataset* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/kodf/README.md) |
82
82
 
83
83
  <div id="data-preparation"></div>
84
84
 
@@ -128,9 +128,9 @@ DATA_ROOT/
128
128
 
129
129
  Click the links below to view the specific preprocessing details for each dataset:
130
130
 
131
- * [**Celeb-DF V2 Preprocess**](/preprocess/celeb_df_v2/README.md)
132
- * [**FaceForensics++ Preprocess**](/preprocess/ff++/README.md)
133
- * [**KoDF Preprocess**](/preprocess/kodf/README.md)
131
+ * [**Celeb-DF V2 Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/celeb_df_v2/README.md)
132
+ * [**FaceForensics++ Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/ff++/README.md)
133
+ * [**KoDF Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/kodf/README.md)
134
134
 
135
135
  ## 🏗 Model Architecture
136
136
 
@@ -138,9 +138,9 @@ Multi Scale Efficient Global Context Vision Transformer is an optimized multi-sc
138
138
 
139
139
  #### Explore More Details
140
140
 
141
- - [Model Architecture: MS-EffViT](deepguard/MS_EffViT.md) - _**Multi Scale Efficient Vision Transformer**_
141
+ - [Model Architecture: MS-EffViT](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/MS_EffViT.md) - _**Multi Scale Efficient Vision Transformer**_
142
142
 
143
- - [Advanced Architecture: MS-EFFGCViT](deepguard/MS_EffGCViT.md) - _**Multi Scale Efficient Global Context Vision Transformer**_
143
+ - [Advanced Architecture: MS-EFFGCViT](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/MS_EffGCViT.md) - _**Multi Scale Efficient Global Context Vision Transformer**_
144
144
 
145
145
 
146
146
  <p align="center">
@@ -155,15 +155,15 @@ We utilizes two distinct types of self-attention to capture both long-range and
155
155
 
156
156
 
157
157
  <p align="center">
158
- <img src=https://raw.githubusercontent.com/HanMoonSub/DeepGuard/main/docs/architectures/window_attention.JPG width="100%" height="300">
158
+ <img src=docs/architectures/window_attention.JPG width="100%" height="300">
159
159
  </p>
160
160
 
161
161
  ## 🧬 Model Zoo
162
162
 
163
163
  | Model | Resolution | # Total Params(M) | # Backbone(M) | # L-ViT(M) | # H-ViT(M) | FLOPs (G) | Model Config |
164
164
  | ----- | ---------- | -------------- | ----------- |------------- | ------------- | -------------- | ------- |
165
- | ⚡ ms_eff_gcvit_b0 | 224 X 224 | 8.7 | 3.6(41.4%) | 1.7(19.5%) | 3.3(37.9%) | 0.87 | [spec](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
166
- | 🔥 ms_eff_gcvit_b5 | 384 X 384 | 50.3 | 27.3(54.3%) | 6.6(13.1%) | 16.1(32.0%) | 13.64 | [spec](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
165
+ | ⚡ ms_eff_gcvit_b0 | 224 X 224 | 8.7 | 3.6(41.4%) | 1.7(19.5%) | 3.3(37.9%) | 0.87 | [spec](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
166
+ | 🔥 ms_eff_gcvit_b5 | 384 X 384 | 50.3 | 27.3(54.3%) | 6.6(13.1%) | 16.1(32.0%) | 13.64 | [spec](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
167
167
 
168
168
  ## 🚀 Training
169
169
 
@@ -205,22 +205,22 @@ We provide training scripts for both `ms_eff_vit` and `ms_eff_gcvit`. We recomme
205
205
 
206
206
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
207
207
  | ------------- | -------- | -------- | ---------- | -------- | ----- |
208
- | ms_eff_gcvit_b0 | 0.9842 | 0.9965 | 0.0283 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_celeb_df_v2.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
209
- | ms_eff_gcvit_b5 | 0.9981 | 0.9984 | 0.0089 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_celeb_df_v2.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
208
+ | ms_eff_gcvit_b0 | 0.9842 | 0.9965 | 0.0283 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_celeb_df_v2.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
209
+ | ms_eff_gcvit_b5 | 0.9981 | 0.9984 | 0.0089 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_celeb_df_v2.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
210
210
 
211
211
  **FaceForensics++ Pretrained Models**
212
212
 
213
213
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
214
214
  | ------------- | -------- | -------- | ---------- | -------- | ------ |
215
- | ms_eff_gcvit_b0 | 0.9808 | 0.9969 | 0.0637| [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_ff++.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
216
- | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_ff++.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
215
+ | ms_eff_gcvit_b0 | 0.9808 | 0.9969 | 0.0637| [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_ff++.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
216
+ | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_ff++.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
217
217
 
218
218
  **KoDF Pretrained Models**
219
219
 
220
220
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
221
221
  | ------------- | -------- | -------- | ---------- | -------- | ------ |
222
- | ms_eff_gcvit_b0 | 0.9655 | 0.9792 | 0.1237 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b0_kodf.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
223
- | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b5_kodf.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
222
+ | ms_eff_gcvit_b0 | 0.9655 | 0.9792 | 0.1237 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b0_kodf.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
223
+ | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b5_kodf.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
224
224
 
225
225
  ## 💻 Model Usage
226
226
 
@@ -473,14 +473,14 @@ result = explainer.display_heatmap_bbox_on_image(
473
473
 
474
474
  The jupyter notebooks themselves can be found under the tutorials folder in the git repository.
475
475
 
476
- - [Notebook tutorial: Predict DeepFake Image with ImagePredictor](./tutorials/predict_image.ipynb)
477
- - [Notebook tutorial: Predict DeepFake Video with VideoPredictor](./tutorials/predict_video.ipynb)
478
- - [Notebook tutorial: Low-Level Branch XAI Visualization (HiResCAM, GradCAMElementWise, LayerCAM)](./tutorials/low_level_visualization.ipynb)
479
- - [Notebook tutorial: High-Level Branch XAI Visualization (EigenGradCAM, GradCAM++, XGradCAM)](./tutorials/high_level_visualization.ipynb)
480
- - [Notebook tutorial: MS-EffViT Low-Level Branch Explainability](./tutorials/ms_eff_vit_low_level.ipynb)
481
- - [Notebook tutorial: MS-EffViT High-Level Branch Explainability](./tutorials/ms_eff_vit_high_level.ipynb)
482
- - [Notebook tutorial: MS-EffGCViT Low-Level Branch Explainability](./tutorials/ms_eff_gcvit_low_level.ipynb)
483
- - [Notebook tutorial: MS-EffGCViT High-Level Branch Explainability](./tutorials/ms_eff_gcvit_high_level.ipynb)
476
+ - [Notebook tutorial: Predict DeepFake Image with ImagePredictor](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/predict_image.ipynb)
477
+ - [Notebook tutorial: Predict DeepFake Video with VideoPredictor](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/predict_video.ipynb)
478
+ - [Notebook tutorial: Low-Level Branch XAI Visualization (HiResCAM, GradCAMElementWise, LayerCAM)](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/low_level_visualization.ipynb)
479
+ - [Notebook tutorial: High-Level Branch XAI Visualization (EigenGradCAM, GradCAM++, XGradCAM)](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/high_level_visualization.ipynb)
480
+ - [Notebook tutorial: MS-EffViT Low-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_vit_low_level.ipynb)
481
+ - [Notebook tutorial: MS-EffViT High-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_vit_high_level.ipynb)
482
+ - [Notebook tutorial: MS-EffGCViT Low-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_gcvit_low_level.ipynb)
483
+ - [Notebook tutorial: MS-EffGCViT High-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_gcvit_high_level.ipynb)
484
484
 
485
485
  ## 📬 Authors
486
486
 
@@ -505,4 +505,4 @@ _**This project was developed as a Senior Graduation Project by the Department o
505
505
 
506
506
  ## ⚖️ License
507
507
 
508
- This project is licensed under the terms of the MIT license.
508
+ This project is licensed under the terms of the MIT license.
{deepguard-0.2.4 → deepguard-0.3.0}/deepguard.egg-info/PKG-INFO RENAMED
@@ -1,6 +1,6 @@
1
1
  Metadata-Version: 2.4
2
2
  Name: deepguard
3
- Version: 0.2.4
3
+ Version: 0.3.0
4
4
  Summary: Multi-Scale Efficient Global Context Vision Transformer for Robust Deepfake Detection
5
5
  Author-email: seoyunje <seoyunje2001@gmail.com>
6
6
  License: MIT
@@ -55,8 +55,8 @@ Dynamic: license-file
55
55
  </p>
56
56
 
57
57
  <p align="center">
58
- <a href="README_KR.md"><b>🇰🇷 한국어 버전</b></a> |
59
- <a href="README_JP.md"><b>🇯🇵 日本語版</b></a> |
58
+ <a href="https://github.com/HanMoonSub/DeepGuard/blob/main/README_KR.md"><b>🇰🇷 한국어 버전</b></a> |
59
+ <a href="https://github.com/HanMoonSub/DeepGuard/blob/main/README_JP.md"><b>🇯🇵 日本語版</b></a> |
60
60
  <a href="#-model-evaluation"><b>📈 Model Evaluation</b></a> |
61
61
  <a href="#-predict-image--video"><b>🔮 Try Demo</b></a>
62
62
  </p>
@@ -102,9 +102,9 @@ To evaluate the generalization and robustness of our deepfake detection model, w
102
102
 
103
103
  | Dataset | Real Videos | Fake Videos | Year | Participants | Description (Paper Title) | Details |
104
104
  | :--- | :---: | :---: | :---: | :---: | :--- | :---: |
105
- | **Celeb-DF-v2** | 890 | 5,639 | 2019 | 59 | *A Large-scale Challenging Dataset for DeepFake Forensics* | [🔗 Readme](preprocess/celeb_df_v2/README.md) |
106
- | **FaceForensics++** | 1,000 | 6,000 | 2019 | 1,000 | *Learning to Detect Manipulated Facial Images* | [🔗 Readme](preprocess/ff++/README.md) |
107
- | **KoDF** | 62,166 | 175,776 | 2020 | 400 | *Large-Scale Korean Deepfake Detection Dataset* | [🔗 Readme](preprocess/kodf/README.md) |
105
+ | **Celeb-DF-v2** | 890 | 5,639 | 2019 | 59 | *A Large-scale Challenging Dataset for DeepFake Forensics* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/celeb_df_v2/README.md) |
106
+ | **FaceForensics++** | 1,000 | 6,000 | 2019 | 1,000 | *Learning to Detect Manipulated Facial Images* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/ff++/README.md) |
107
+ | **KoDF** | 62,166 | 175,776 | 2020 | 400 | *Large-Scale Korean Deepfake Detection Dataset* | [🔗 Readme](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/kodf/README.md) |
108
108
 
109
109
  <div id="data-preparation"></div>
110
110
 
@@ -154,9 +154,9 @@ DATA_ROOT/
154
154
 
155
155
  Click the links below to view the specific preprocessing details for each dataset:
156
156
 
157
- * [**Celeb-DF V2 Preprocess**](/preprocess/celeb_df_v2/README.md)
158
- * [**FaceForensics++ Preprocess**](/preprocess/ff++/README.md)
159
- * [**KoDF Preprocess**](/preprocess/kodf/README.md)
157
+ * [**Celeb-DF V2 Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/celeb_df_v2/README.md)
158
+ * [**FaceForensics++ Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/ff++/README.md)
159
+ * [**KoDF Preprocess**](https://github.com/HanMoonSub/DeepGuard/blob/main/preprocess/kodf/README.md)
160
160
 
161
161
  ## 🏗 Model Architecture
162
162
 
@@ -164,9 +164,9 @@ Multi Scale Efficient Global Context Vision Transformer is an optimized multi-sc
164
164
 
165
165
  #### Explore More Details
166
166
 
167
- - [Model Architecture: MS-EffViT](deepguard/MS_EffViT.md) - _**Multi Scale Efficient Vision Transformer**_
167
+ - [Model Architecture: MS-EffViT](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/MS_EffViT.md) - _**Multi Scale Efficient Vision Transformer**_
168
168
 
169
- - [Advanced Architecture: MS-EFFGCViT](deepguard/MS_EffGCViT.md) - _**Multi Scale Efficient Global Context Vision Transformer**_
169
+ - [Advanced Architecture: MS-EFFGCViT](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/MS_EffGCViT.md) - _**Multi Scale Efficient Global Context Vision Transformer**_
170
170
 
171
171
 
172
172
  <p align="center">
@@ -181,15 +181,15 @@ We utilizes two distinct types of self-attention to capture both long-range and
181
181
 
182
182
 
183
183
  <p align="center">
184
- <img src=https://raw.githubusercontent.com/HanMoonSub/DeepGuard/main/docs/architectures/window_attention.JPG width="100%" height="300">
184
+ <img src=docs/architectures/window_attention.JPG width="100%" height="300">
185
185
  </p>
186
186
 
187
187
  ## 🧬 Model Zoo
188
188
 
189
189
  | Model | Resolution | # Total Params(M) | # Backbone(M) | # L-ViT(M) | # H-ViT(M) | FLOPs (G) | Model Config |
190
190
  | ----- | ---------- | -------------- | ----------- |------------- | ------------- | -------------- | ------- |
191
- | ⚡ ms_eff_gcvit_b0 | 224 X 224 | 8.7 | 3.6(41.4%) | 1.7(19.5%) | 3.3(37.9%) | 0.87 | [spec](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
192
- | 🔥 ms_eff_gcvit_b5 | 384 X 384 | 50.3 | 27.3(54.3%) | 6.6(13.1%) | 16.1(32.0%) | 13.64 | [spec](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
191
+ | ⚡ ms_eff_gcvit_b0 | 224 X 224 | 8.7 | 3.6(41.4%) | 1.7(19.5%) | 3.3(37.9%) | 0.87 | [spec](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
192
+ | 🔥 ms_eff_gcvit_b5 | 384 X 384 | 50.3 | 27.3(54.3%) | 6.6(13.1%) | 16.1(32.0%) | 13.64 | [spec](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
193
193
 
194
194
  ## 🚀 Training
195
195
 
@@ -231,22 +231,22 @@ We provide training scripts for both `ms_eff_vit` and `ms_eff_gcvit`. We recomme
231
231
 
232
232
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
233
233
  | ------------- | -------- | -------- | ---------- | -------- | ----- |
234
- | ms_eff_gcvit_b0 | 0.9842 | 0.9965 | 0.0283 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_celeb_df_v2.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
235
- | ms_eff_gcvit_b5 | 0.9981 | 0.9984 | 0.0089 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_celeb_df_v2.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
234
+ | ms_eff_gcvit_b0 | 0.9842 | 0.9965 | 0.0283 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_celeb_df_v2.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
235
+ | ms_eff_gcvit_b5 | 0.9981 | 0.9984 | 0.0089 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_celeb_df_v2.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
236
236
 
237
237
  **FaceForensics++ Pretrained Models**
238
238
 
239
239
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
240
240
  | ------------- | -------- | -------- | ---------- | -------- | ------ |
241
- | ms_eff_gcvit_b0 | 0.9808 | 0.9969 | 0.0637| [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_ff++.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
242
- | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_ff++.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
241
+ | ms_eff_gcvit_b0 | 0.9808 | 0.9969 | 0.0637| [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b0_ff++.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
242
+ | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.1.0/ms_eff_gcvit_b5_ff++.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
243
243
 
244
244
  **KoDF Pretrained Models**
245
245
 
246
246
  | Model Variant | Test@Acc | Test@Auc | Test@log_loss | Download | Train Config |
247
247
  | ------------- | -------- | -------- | ---------- | -------- | ------ |
248
- | ms_eff_gcvit_b0 | 0.9655 | 0.9792 | 0.1237 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b0_kodf.bin) | [recipe](deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
249
- | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b5_kodf.bin) | [recipe](deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
248
+ | ms_eff_gcvit_b0 | 0.9655 | 0.9792 | 0.1237 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b0_kodf.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b0/celeb_df_v2.yaml) |
249
+ | ms_eff_gcvit_b5 | 0.9850 | 0.9974 | 0.0492 | [model](https://github.com/HanMoonSub/DeepGuard/releases/download/v0.2.0/ms_eff_gcvit_b5_kodf.bin) | [recipe](https://github.com/HanMoonSub/DeepGuard/blob/main/deepguard/config/ms_eff_gcvit_b5/celeb_df_v2.yaml) |
250
250
 
251
251
  ## 💻 Model Usage
252
252
 
@@ -499,14 +499,14 @@ result = explainer.display_heatmap_bbox_on_image(
499
499
 
500
500
  The jupyter notebooks themselves can be found under the tutorials folder in the git repository.
501
501
 
502
- - [Notebook tutorial: Predict DeepFake Image with ImagePredictor](./tutorials/predict_image.ipynb)
503
- - [Notebook tutorial: Predict DeepFake Video with VideoPredictor](./tutorials/predict_video.ipynb)
504
- - [Notebook tutorial: Low-Level Branch XAI Visualization (HiResCAM, GradCAMElementWise, LayerCAM)](./tutorials/low_level_visualization.ipynb)
505
- - [Notebook tutorial: High-Level Branch XAI Visualization (EigenGradCAM, GradCAM++, XGradCAM)](./tutorials/high_level_visualization.ipynb)
506
- - [Notebook tutorial: MS-EffViT Low-Level Branch Explainability](./tutorials/ms_eff_vit_low_level.ipynb)
507
- - [Notebook tutorial: MS-EffViT High-Level Branch Explainability](./tutorials/ms_eff_vit_high_level.ipynb)
508
- - [Notebook tutorial: MS-EffGCViT Low-Level Branch Explainability](./tutorials/ms_eff_gcvit_low_level.ipynb)
509
- - [Notebook tutorial: MS-EffGCViT High-Level Branch Explainability](./tutorials/ms_eff_gcvit_high_level.ipynb)
502
+ - [Notebook tutorial: Predict DeepFake Image with ImagePredictor](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/predict_image.ipynb)
503
+ - [Notebook tutorial: Predict DeepFake Video with VideoPredictor](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/predict_video.ipynb)
504
+ - [Notebook tutorial: Low-Level Branch XAI Visualization (HiResCAM, GradCAMElementWise, LayerCAM)](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/low_level_visualization.ipynb)
505
+ - [Notebook tutorial: High-Level Branch XAI Visualization (EigenGradCAM, GradCAM++, XGradCAM)](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/high_level_visualization.ipynb)
506
+ - [Notebook tutorial: MS-EffViT Low-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_vit_low_level.ipynb)
507
+ - [Notebook tutorial: MS-EffViT High-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_vit_high_level.ipynb)
508
+ - [Notebook tutorial: MS-EffGCViT Low-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_gcvit_low_level.ipynb)
509
+ - [Notebook tutorial: MS-EffGCViT High-Level Branch Explainability](https://github.com/HanMoonSub/DeepGuard/blob/main/tutorials/ms_eff_gcvit_high_level.ipynb)
510
510
 
511
511
  ## 📬 Authors
512
512
 
{deepguard-0.2.4 → deepguard-0.3.0}/explainability/explainer/base_explainer.py RENAMED
@@ -18,6 +18,7 @@ class BaseExplainer:
18
18
  branch_level: Literal["low", "high"] = "high",
19
19
  l_stage_idx: int = -1, # gcvit low branch stage index (0~3)
20
20
  block_idx: int = -1, # vit, gcvit block index
21
+ model: Optional[torch.nn.Module] = None,
21
22
  ):
22
23
 
23
24
  self.device = "cuda:0" if torch.cuda.is_available() else 'cpu'
@@ -34,7 +35,9 @@ class BaseExplainer:
34
35
  self.transformer_type = model_name.split("_")[-2] # vit, gcvit
35
36
 
36
37
  # Model & Tools setup
37
- self.model = timm.create_model(model_name, pretrained=True, dataset=dataset)
38
+ self.model = model if model is not None else timm.create_model(
39
+ model_name, pretrained=True, dataset=dataset
40
+ )
38
41
  self.face_detector = FaceDetector2(conf_thres)
39
42
  self.img_size = [224,224] if self.model_variant == "b0" else [384,384]
40
43
  self.transforms = get_test_transforms(img_size=self.img_size)
{deepguard-0.2.4 → deepguard-0.3.0}/pyproject.toml RENAMED
@@ -8,7 +8,7 @@ build-backend = "setuptools.build_meta"
8
8
  # ─────────────────────────────────────────────
9
9
  [project]
10
10
  name = "deepguard" # pip install deepguard 할 때 이 이름
11
- version = "0.2.4" # pip install deepguard==0.2.3 으로 버전 고정 가능
11
+ version = "0.3.0" # pip install deepguard==0.2.4 으로 버전 고정 가능
12
12
  description = "Multi-Scale Efficient Global Context Vision Transformer for Robust Deepfake Detection"
13
13
  readme = "README.md" # PyPI 페이지에 표시될 설명 (마크다운)
14
14
  requires-python = ">=3.10" # 이 버전 미만이면 pip install 자체를 막음
{deepguard-0.2.4 → deepguard-0.3.0}/LICENSE RENAMED
File without changes
{deepguard-0.2.4 → deepguard-0.3.0}/inference/utils.py RENAMED
File without changes
{deepguard-0.2.4 → deepguard-0.3.0}/setup.cfg RENAMED
File without changes