civicbench 0.1.0__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.
@@ -0,0 +1,202 @@
1
+
2
+ Apache License
3
+ Version 2.0, January 2004
4
+ http://www.apache.org/licenses/
5
+
6
+ TERMS AND CONDITIONS FOR USE, REPRODUCTION, AND DISTRIBUTION
7
+
8
+ 1. Definitions.
9
+
10
+ "License" shall mean the terms and conditions for use, reproduction,
11
+ and distribution as defined by Sections 1 through 9 of this document.
12
+
13
+ "Licensor" shall mean the copyright owner or entity authorized by
14
+ the copyright owner that is granting the License.
15
+
16
+ "Legal Entity" shall mean the union of the acting entity and all
17
+ other entities that control, are controlled by, or are under common
18
+ control with that entity. For the purposes of this definition,
19
+ "control" means (i) the power, direct or indirect, to cause the
20
+ direction or management of such entity, whether by contract or
21
+ otherwise, or (ii) ownership of fifty percent (50%) or more of the
22
+ outstanding shares, or (iii) beneficial ownership of such entity.
23
+
24
+ "You" (or "Your") shall mean an individual or Legal Entity
25
+ exercising permissions granted by this License.
26
+
27
+ "Source" form shall mean the preferred form for making modifications,
28
+ including but not limited to software source code, documentation
29
+ source, and configuration files.
30
+
31
+ "Object" form shall mean any form resulting from mechanical
32
+ transformation or translation of a Source form, including but
33
+ not limited to compiled object code, generated documentation,
34
+ and conversions to other media types.
35
+
36
+ "Work" shall mean the work of authorship, whether in Source or
37
+ Object form, made available under the License, as indicated by a
38
+ copyright notice that is included in or attached to the work
39
+ (an example is provided in the Appendix below).
40
+
41
+ "Derivative Works" shall mean any work, whether in Source or Object
42
+ form, that is based on (or derived from) the Work and for which the
43
+ editorial revisions, annotations, elaborations, or other modifications
44
+ represent, as a whole, an original work of authorship. For the purposes
45
+ of this License, Derivative Works shall not include works that remain
46
+ separable from, or merely link (or bind by name) to the interfaces of,
47
+ the Work and Derivative Works thereof.
48
+
49
+ "Contribution" shall mean any work of authorship, including
50
+ the original version of the Work and any modifications or additions
51
+ to that Work or Derivative Works thereof, that is intentionally
52
+ submitted to Licensor for inclusion in the Work by the copyright owner
53
+ or by an individual or Legal Entity authorized to submit on behalf of
54
+ the copyright owner. For the purposes of this definition, "submitted"
55
+ means any form of electronic, verbal, or written communication sent
56
+ to the Licensor or its representatives, including but not limited to
57
+ communication on electronic mailing lists, source code control systems,
58
+ and issue tracking systems that are managed by, or on behalf of, the
59
+ Licensor for the purpose of discussing and improving the Work, but
60
+ excluding communication that is conspicuously marked or otherwise
61
+ designated in writing by the copyright owner as "Not a Contribution."
62
+
63
+ "Contributor" shall mean Licensor and any individual or Legal Entity
64
+ on behalf of whom a Contribution has been received by Licensor and
65
+ subsequently incorporated within the Work.
66
+
67
+ 2. Grant of Copyright License. Subject to the terms and conditions of
68
+ this License, each Contributor hereby grants to You a perpetual,
69
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
70
+ copyright license to reproduce, prepare Derivative Works of,
71
+ publicly display, publicly perform, sublicense, and distribute the
72
+ Work and such Derivative Works in Source or Object form.
73
+
74
+ 3. Grant of Patent License. Subject to the terms and conditions of
75
+ this License, each Contributor hereby grants to You a perpetual,
76
+ worldwide, non-exclusive, no-charge, royalty-free, irrevocable
77
+ (except as stated in this section) patent license to make, have made,
78
+ use, offer to sell, sell, import, and otherwise transfer the Work,
79
+ where such license applies only to those patent claims licensable
80
+ by such Contributor that are necessarily infringed by their
81
+ Contribution(s) alone or by combination of their Contribution(s)
82
+ with the Work to which such Contribution(s) was submitted. If You
83
+ institute patent litigation against any entity (including a
84
+ cross-claim or counterclaim in a lawsuit) alleging that the Work
85
+ or a Contribution incorporated within the Work constitutes direct
86
+ or contributory patent infringement, then any patent licenses
87
+ granted to You under this License for that Work shall terminate
88
+ as of the date such litigation is filed.
89
+
90
+ 4. Redistribution. You may reproduce and distribute copies of the
91
+ Work or Derivative Works thereof in any medium, with or without
92
+ modifications, and in Source or Object form, provided that You
93
+ meet the following conditions:
94
+
95
+ (a) You must give any other recipients of the Work or
96
+ Derivative Works a copy of this License; and
97
+
98
+ (b) You must cause any modified files to carry prominent notices
99
+ stating that You changed the files; and
100
+
101
+ (c) You must retain, in the Source form of any Derivative Works
102
+ that You distribute, all copyright, patent, trademark, and
103
+ attribution notices from the Source form of the Work,
104
+ excluding those notices that do not pertain to any part of
105
+ the Derivative Works; and
106
+
107
+ (d) If the Work includes a "NOTICE" text file as part of its
108
+ distribution, then any Derivative Works that You distribute must
109
+ include a readable copy of the attribution notices contained
110
+ within such NOTICE file, excluding those notices that do not
111
+ pertain to any part of the Derivative Works, in at least one
112
+ of the following places: within a NOTICE text file distributed
113
+ as part of the Derivative Works; within the Source form or
114
+ documentation, if provided along with the Derivative Works; or,
115
+ within a display generated by the Derivative Works, if and
116
+ wherever such third-party notices normally appear. The contents
117
+ of the NOTICE file are for informational purposes only and
118
+ do not modify the License. You may add Your own attribution
119
+ notices within Derivative Works that You distribute, alongside
120
+ or as an addendum to the NOTICE text from the Work, provided
121
+ that such additional attribution notices cannot be construed
122
+ as modifying the License.
123
+
124
+ You may add Your own copyright statement to Your modifications and
125
+ may provide additional or different license terms and conditions
126
+ for use, reproduction, or distribution of Your modifications, or
127
+ for any such Derivative Works as a whole, provided Your use,
128
+ reproduction, and distribution of the Work otherwise complies with
129
+ the conditions stated in this License.
130
+
131
+ 5. Submission of Contributions. Unless You explicitly state otherwise,
132
+ any Contribution intentionally submitted for inclusion in the Work
133
+ by You to the Licensor shall be under the terms and conditions of
134
+ this License, without any additional terms or conditions.
135
+ Notwithstanding the above, nothing herein shall supersede or modify
136
+ the terms of any separate license agreement you may have executed
137
+ with Licensor regarding such Contributions.
138
+
139
+ 6. Trademarks. This License does not grant permission to use the trade
140
+ names, trademarks, service marks, or product names of the Licensor,
141
+ except as required for reasonable and customary use in describing the
142
+ origin of the Work and reproducing the content of the NOTICE file.
143
+
144
+ 7. Disclaimer of Warranty. Unless required by applicable law or
145
+ agreed to in writing, Licensor provides the Work (and each
146
+ Contributor provides its Contributions) on an "AS IS" BASIS,
147
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or
148
+ implied, including, without limitation, any warranties or conditions
149
+ of TITLE, NON-INFRINGEMENT, MERCHANTABILITY, or FITNESS FOR A
150
+ PARTICULAR PURPOSE. You are solely responsible for determining the
151
+ appropriateness of using or redistributing the Work and assume any
152
+ risks associated with Your exercise of permissions under this License.
153
+
154
+ 8. Limitation of Liability. In no event and under no legal theory,
155
+ whether in tort (including negligence), contract, or otherwise,
156
+ unless required by applicable law (such as deliberate and grossly
157
+ negligent acts) or agreed to in writing, shall any Contributor be
158
+ liable to You for damages, including any direct, indirect, special,
159
+ incidental, or consequential damages of any character arising as a
160
+ result of this License or out of the use or inability to use the
161
+ Work (including but not limited to damages for loss of goodwill,
162
+ work stoppage, computer failure or malfunction, or any and all
163
+ other commercial damages or losses), even if such Contributor
164
+ has been advised of the possibility of such damages.
165
+
166
+ 9. Accepting Warranty or Additional Liability. While redistributing
167
+ the Work or Derivative Works thereof, You may choose to offer,
168
+ and charge a fee for, acceptance of support, warranty, indemnity,
169
+ or other liability obligations and/or rights consistent with this
170
+ License. However, in accepting such obligations, You may act only
171
+ on Your own behalf and on Your sole responsibility, not on behalf
172
+ of any other Contributor, and only if You agree to indemnify,
173
+ defend, and hold each Contributor harmless for any liability
174
+ incurred by, or claims asserted against, such Contributor by reason
175
+ of your accepting any such warranty or additional liability.
176
+
177
+ END OF TERMS AND CONDITIONS
178
+
179
+ APPENDIX: How to apply the Apache License to your work.
180
+
181
+ To apply the Apache License to your work, attach the following
182
+ boilerplate notice, with the fields enclosed by brackets "[]"
183
+ replaced with your own identifying information. (Don't include
184
+ the brackets!) The text should be enclosed in the appropriate
185
+ comment syntax for the file format. We also recommend that a
186
+ file or class name and description of purpose be included on the
187
+ same "printed page" as the copyright notice for easier
188
+ identification within third-party archives.
189
+
190
+ Copyright [yyyy] [name of copyright owner]
191
+
192
+ Licensed under the Apache License, Version 2.0 (the "License");
193
+ you may not use this file except in compliance with the License.
194
+ You may obtain a copy of the License at
195
+
196
+ http://www.apache.org/licenses/LICENSE-2.0
197
+
198
+ Unless required by applicable law or agreed to in writing, software
199
+ distributed under the License is distributed on an "AS IS" BASIS,
200
+ WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
201
+ See the License for the specific language governing permissions and
202
+ limitations under the License.
@@ -0,0 +1,2 @@
1
+ # The tests need tests/_toy.py and tau_anchor/ (repository only), so the sdist ships the package alone.
2
+ prune tests
@@ -0,0 +1,4 @@
1
+ CivicBench
2
+ Copyright 2026 The CivicBench authors
3
+
4
+ This product is licensed under the Apache License, Version 2.0 (see LICENSE).
@@ -0,0 +1,230 @@
1
+ Metadata-Version: 2.4
2
+ Name: civicbench
3
+ Version: 0.1.0
4
+ Summary: Load CivicBench and attack your own video detector with the benchmark's protocol (PGD, APGD and Square on the 8-bit grid).
5
+ Author: Mert Onur Cakiroglu, Elham Buxton, Mehmet Dalkilic, Hasan Kurban
6
+ License-Expression: Apache-2.0
7
+ Project-URL: Homepage, https://github.com/KurbanIntelligenceLab/CivicBench
8
+ Project-URL: Documentation, https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/docs/toolkit.md
9
+ Project-URL: Dataset, https://huggingface.co/datasets/kurbanintelligencelab/CivicBench
10
+ Project-URL: Issues, https://github.com/KurbanIntelligenceLab/CivicBench/issues
11
+ Keywords: video,deepfake detection,adversarial robustness,benchmark,generated video
12
+ Classifier: Development Status :: 4 - Beta
13
+ Classifier: Intended Audience :: Science/Research
14
+ Classifier: Operating System :: OS Independent
15
+ Classifier: Programming Language :: Python :: 3
16
+ Classifier: Programming Language :: Python :: 3.10
17
+ Classifier: Programming Language :: Python :: 3.11
18
+ Classifier: Programming Language :: Python :: 3.12
19
+ Classifier: Programming Language :: Python :: 3.13
20
+ Classifier: Topic :: Multimedia :: Video
21
+ Classifier: Topic :: Scientific/Engineering :: Artificial Intelligence
22
+ Requires-Python: >=3.10
23
+ Description-Content-Type: text/markdown
24
+ License-File: LICENSE
25
+ License-File: NOTICE
26
+ Requires-Dist: numpy>=1.21
27
+ Requires-Dist: torch>=1.13
28
+ Provides-Extra: data
29
+ Requires-Dist: pyarrow>=10; extra == "data"
30
+ Requires-Dist: Pillow>=9; extra == "data"
31
+ Requires-Dist: datasets>=2.14; extra == "data"
32
+ Provides-Extra: fancy
33
+ Requires-Dist: rich>=13; extra == "fancy"
34
+ Provides-Extra: test
35
+ Requires-Dist: pytest>=7; extra == "test"
36
+ Provides-Extra: tau-anchor
37
+ Requires-Dist: torch>=2.0; extra == "tau-anchor"
38
+ Requires-Dist: huggingface_hub>=0.20; extra == "tau-anchor"
39
+ Provides-Extra: tau-anchor-train
40
+ Requires-Dist: torch>=2.0; extra == "tau-anchor-train"
41
+ Requires-Dist: huggingface_hub>=0.20; extra == "tau-anchor-train"
42
+ Requires-Dist: scikit-learn>=1.0; extra == "tau-anchor-train"
43
+ Requires-Dist: av>=10; extra == "tau-anchor-train"
44
+ Requires-Dist: Pillow>=9; extra == "tau-anchor-train"
45
+ Dynamic: license-file
46
+
47
+ <p align="center">
48
+ <picture>
49
+ <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/banner-dark.jpg">
50
+ <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/banner-light.jpg">
51
+ <img alt="CivicBench. Hiding fakes. Framing real footage." src="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/banner-light.jpg" width="100%">
52
+ </picture>
53
+ </p>
54
+
55
+ <h3 align="center">CivicBench: Robust Detection of AI-Generated Political Video</h3>
56
+
57
+ <p align="center">
58
+ <b>Mert Onur Cakiroglu</b><sup>1</sup> &nbsp;·&nbsp;
59
+ <b>Elham Buxton</b><sup>2</sup> &nbsp;·&nbsp;
60
+ <b>Mehmet Dalkilic</b><sup>1</sup> &nbsp;·&nbsp;
61
+ <b>Hasan Kurban</b><sup>3,*</sup>
62
+ <br>
63
+ <sup>1</sup>Indiana University Bloomington &nbsp;&nbsp;
64
+ <sup>2</sup>University of Illinois Springfield &nbsp;&nbsp;
65
+ <sup>3</sup>Hamad Bin Khalifa University
66
+ <br>
67
+ <sub><sup>*</sup>Corresponding author &nbsp;·&nbsp; Kurban Intelligence Lab</sub>
68
+ </p>
69
+
70
+ <p align="center">
71
+ <a href="#citation"><img alt="Paper" src="https://img.shields.io/badge/Paper-arXiv-B31B1B?style=flat-square&logo=arxiv&logoColor=white"></a>
72
+ <a href="https://huggingface.co/datasets/kurbanintelligencelab/CivicBench"><img alt="Dataset" src="https://img.shields.io/badge/Dataset-Hugging%20Face-FFD21E?style=flat-square&logo=huggingface&logoColor=black"></a>
73
+ <a href="https://pypi.org/project/civicbench/"><img alt="PyPI" src="https://img.shields.io/badge/pip%20install-civicbench-3775A9?style=flat-square&logo=pypi&logoColor=white"></a>
74
+ <a href="https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Apache%202.0-0A7BBB?style=flat-square"></a>
75
+ <img alt="Python" src="https://img.shields.io/badge/Python-3.10%2B-3776AB?style=flat-square&logo=python&logoColor=white">
76
+ </p>
77
+
78
+ <p align="center">
79
+ <a href="#quick-start"><b>Quick start</b></a> &nbsp;·&nbsp;
80
+ <a href="#data"><b>Data</b></a> &nbsp;·&nbsp;
81
+ <a href="https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/docs/toolkit.md"><b>Toolkit guide</b></a> &nbsp;·&nbsp;
82
+ <a href="#leaderboard"><b>Leaderboard</b></a> &nbsp;·&nbsp;
83
+ <a href="https://github.com/KurbanIntelligenceLab/CivicBench/tree/main/tau_anchor"><b>Tau-Anchor</b></a> &nbsp;·&nbsp;
84
+ <a href="#citation"><b>Citation</b></a>
85
+ </p>
86
+
87
+ <br>
88
+
89
+ **CivicBench** tests detectors of AI-generated political video against the two attacks that matter in an election.
90
+ An attacker can change each pixel of a fake by a gray level or two, so that the detector passes it as real
91
+ (**evasion**). The same small change can make the detector flag real footage as generated (**framing**). Anyone who
92
+ wants to dismiss a real video can then point to the detector's verdict.
93
+
94
+ CivicBench measures both, on clips that can be shared, and one command runs it on your own detector.
95
+
96
+ <p align="center">
97
+ <img alt="Attacks on political video in both directions" src="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/teaser.jpg" width="100%">
98
+ </p>
99
+
100
+ ## Highlights
101
+
102
+ - **Two-sided by design.** Every detector is attacked in both directions, at the threshold where it is deployed.
103
+ At one gray level, four published detectors lose every fake they flagged and flag nearly every attacked real clip.
104
+ - **Political and shareable.** 10,800 clips of invented politicians, parties and countries, from 8 open generators
105
+ and 1,350 prompts across 9 kinds of civic scenes, next to 2,632 real clips under open licenses.
106
+ - **One command.** `civicbench evaluate` runs PGD, APGD and Square at any budget and prints the paper's table.
107
+ Every attacked clip is checked after rounding to 8 bits, as a viewer would see it.
108
+ - **Predictive.** A cheap attack of one gray level predicts how many clips keep their label at four.
109
+
110
+ ## Install
111
+
112
+ ```bash
113
+ pip install civicbench # core: numpy and torch
114
+ pip install "civicbench[data]" # + loading from Hugging Face
115
+ pip install "civicbench[fancy]" # + the rich progress display
116
+ ```
117
+
118
+ Python 3.10 or newer.
119
+
120
+ ## Quick start
121
+
122
+ ```python
123
+ import civicbench
124
+
125
+ clips = civicbench.load() # official test inputs and labels
126
+ results = civicbench.evaluate(my_detector, clips) # AUC, TPR, evasion and framing at 1, 2 and 4 gray levels
127
+ ```
128
+
129
+ Or from the command line, at any budgets:
130
+
131
+ ```bash
132
+ civicbench evaluate --detector my_lab.detector:build --budgets 1,2,4,8,16 --out results.json
133
+ ```
134
+
135
+ <p align="center">
136
+ <img alt="civicbench in a terminal" src="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/terminal.png" width="100%">
137
+ </p>
138
+
139
+ <sub>Output of <code>examples/demo_toy.py</code>: a toy detector on synthetic clips, so the numbers carry no meaning.
140
+ It runs in under a minute with no downloads.</sub>
141
+
142
+ ### Bring your own detector
143
+
144
+ A detector is any callable that maps frames to one score.
145
+
146
+ | | |
147
+ |---|---|
148
+ | **Input** | `torch.float32` tensor `[T, 3, H, W]`, RGB in [0, 1]. The official input is 11 frames of 512 x 512. |
149
+ | **Output** | One score. Higher means generated. A clip is flagged when its score reaches the threshold. |
150
+ | **Gradients** | PGD and APGD need a differentiable score. Square needs scores only. |
151
+
152
+ The [toolkit guide](https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/docs/toolkit.md) covers single-clip attacks, custom budgets, the survival prediction and every
153
+ difference from the paper's code.
154
+
155
+ ## Data
156
+
157
+ One open dataset, [kurbanintelligencelab/CivicBench](https://huggingface.co/datasets/kurbanintelligencelab/CivicBench), with no login and no form.
158
+
159
+ | | Clips | License |
160
+ |---|---|---|
161
+ | **Real** | 2,632 clips of briefings, speeches, legislatures, and protests | Each clip's own: public domain, CC BY or CC BY-SA |
162
+ | **Generated** | 10,800 clips from eight open text-to-video generators, with prompts, seeds, and settings | Research on detection, with each generator's terms (see the dataset card) |
163
+
164
+ Each real clip lists its creator, license, source page, start time and the changes we made. The official test input is
165
+ 11 frames of 512 x 512 per clip, stored as lossless PNG with checksums.
166
+
167
+ ## Leaderboard
168
+
169
+ CivicBench at the threshold that flags 5% of real clips. Evasion is the TPR under attack, higher is better. Framing is
170
+ the false-alarm rate on attacked real clips, lower is better. Budgets are in gray levels on the 0 to 255 scale.
171
+
172
+ | Detector | AUC | TPR | Evasion 1 | Evasion 2 | Evasion 4 | Framing 1 | Framing 2 | Framing 4 |
173
+ |:---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
174
+ | **Tau-Anchor** (ours) | **.941** | **.779** | **.731** | **.668** | **.585** | .124 | .141 | .222 |
175
+ | Mavali et al., R4 | .774 | .394 | .339 | .289 | .239 | .217 | .304 | .623 |
176
+ | Mavali et al., R2 | .768 | .381 | .236 | .164 | .135 | .261 | .536 | .884 |
177
+ | Mavali et al., ΔCLIP | .638 | .182 | .135 | .130 | .099 | **.046** | **.046** | **.101** |
178
+ | MV-MobileNet | .587 | .086 | .045 | .043 | .035 | .348 | .493 | .565 |
179
+ | STREAM-MV | .602 | .111 | .018 | .013 | .011 | .768 | .783 | .812 |
180
+ | Shahjahan et al. | .709 | .233 | .023 | .000 | .000 | .913 | .971 | .986 |
181
+ | CLIP probe | .781 | .364 | .000 | .000 | .000 | 1.000 | 1.000 | 1.000 |
182
+ | STREAM | .714 | .309 | .000 | .000 | .000 | 1.000 | 1.000 | 1.000 |
183
+ | D3 | .475 | .028 | .000 | .000 | .000 | 1.000 | 1.000 | 1.000 |
184
+ | ReStraV | .448 | .009 | .000 | .000 | .000 | .986 | 1.000 | 1.000 |
185
+
186
+ Sorted by TPR under evasion at 2 gray levels, then at 1, then by AUC. The paper adds confidence intervals, two general benchmarks and
187
+ external political footage. To add a detector, open a pull request with the `results.json` that
188
+ `civicbench evaluate --out` writes.
189
+
190
+ ## Tau-Anchor
191
+
192
+ Tau-Anchor is the detector from the paper. It fine-tunes an image encoder that was pre-trained on attacked images,
193
+ attacks real and generated clips during training, and adds a margin term for the fakes nearest its deployed threshold.
194
+ Its code, exact training recipe and model card are in [`tau_anchor/`](https://github.com/KurbanIntelligenceLab/CivicBench/tree/main/tau_anchor). The weights are on
195
+ Hugging Face at [`kurbanintelligencelab/Tau-Anchor`](https://huggingface.co/kurbanintelligencelab/Tau-Anchor).
196
+
197
+ ## Repository layout
198
+
199
+ ```
200
+ src/civicbench/ the pip package: load, attack, evaluate, predict_survival, CLI
201
+ tests/ unit tests and a golden test against the paper's attack code
202
+ examples/ demo_toy.py, a one-minute end-to-end run with no downloads
203
+ tau_anchor/ Tau-Anchor training and inference
204
+ docs/ toolkit guide and assets
205
+ ```
206
+
207
+ ## Responsible use
208
+
209
+ The release contains no attacked real clips, and thus no ready-made forgeries of real footage. The generated clips show invented people. If a clip resembles a real person, report it
210
+ through the contact on the dataset card and we will remove it.
211
+
212
+ ## Citation
213
+
214
+ ```bibtex
215
+ @misc{cakiroglu2026civicbench,
216
+ title = {CivicBench: Robust Detection of AI-Generated Political Video},
217
+ author = {Cakiroglu, Mert Onur and Buxton, Elham and Dalkilic, Mehmet and Kurban, Hasan},
218
+ year = {2026}
219
+ }
220
+ ```
221
+
222
+ ## License
223
+
224
+ The code is released under the [Apache License 2.0](https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/LICENSE). The data follow the licenses in the table above and on
225
+ each dataset card. Generated clips also follow the terms of the generator that rendered them.
226
+
227
+ ## Acknowledgements
228
+
229
+ We thank the creators of the openly licensed footage, credited clip by clip in the dataset, and the developers of
230
+ CogVideoX1.5, HunyuanVideo-1.5, Kandinsky 5, LTX-Video, Mochi, SANA-Video, Wan2.1 and Wan2.2.
@@ -0,0 +1,184 @@
1
+ <p align="center">
2
+ <picture>
3
+ <source media="(prefers-color-scheme: dark)" srcset="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/banner-dark.jpg">
4
+ <source media="(prefers-color-scheme: light)" srcset="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/banner-light.jpg">
5
+ <img alt="CivicBench. Hiding fakes. Framing real footage." src="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/banner-light.jpg" width="100%">
6
+ </picture>
7
+ </p>
8
+
9
+ <h3 align="center">CivicBench: Robust Detection of AI-Generated Political Video</h3>
10
+
11
+ <p align="center">
12
+ <b>Mert Onur Cakiroglu</b><sup>1</sup> &nbsp;·&nbsp;
13
+ <b>Elham Buxton</b><sup>2</sup> &nbsp;·&nbsp;
14
+ <b>Mehmet Dalkilic</b><sup>1</sup> &nbsp;·&nbsp;
15
+ <b>Hasan Kurban</b><sup>3,*</sup>
16
+ <br>
17
+ <sup>1</sup>Indiana University Bloomington &nbsp;&nbsp;
18
+ <sup>2</sup>University of Illinois Springfield &nbsp;&nbsp;
19
+ <sup>3</sup>Hamad Bin Khalifa University
20
+ <br>
21
+ <sub><sup>*</sup>Corresponding author &nbsp;·&nbsp; Kurban Intelligence Lab</sub>
22
+ </p>
23
+
24
+ <p align="center">
25
+ <a href="#citation"><img alt="Paper" src="https://img.shields.io/badge/Paper-arXiv-B31B1B?style=flat-square&logo=arxiv&logoColor=white"></a>
26
+ <a href="https://huggingface.co/datasets/kurbanintelligencelab/CivicBench"><img alt="Dataset" src="https://img.shields.io/badge/Dataset-Hugging%20Face-FFD21E?style=flat-square&logo=huggingface&logoColor=black"></a>
27
+ <a href="https://pypi.org/project/civicbench/"><img alt="PyPI" src="https://img.shields.io/badge/pip%20install-civicbench-3775A9?style=flat-square&logo=pypi&logoColor=white"></a>
28
+ <a href="https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/LICENSE"><img alt="License" src="https://img.shields.io/badge/License-Apache%202.0-0A7BBB?style=flat-square"></a>
29
+ <img alt="Python" src="https://img.shields.io/badge/Python-3.10%2B-3776AB?style=flat-square&logo=python&logoColor=white">
30
+ </p>
31
+
32
+ <p align="center">
33
+ <a href="#quick-start"><b>Quick start</b></a> &nbsp;·&nbsp;
34
+ <a href="#data"><b>Data</b></a> &nbsp;·&nbsp;
35
+ <a href="https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/docs/toolkit.md"><b>Toolkit guide</b></a> &nbsp;·&nbsp;
36
+ <a href="#leaderboard"><b>Leaderboard</b></a> &nbsp;·&nbsp;
37
+ <a href="https://github.com/KurbanIntelligenceLab/CivicBench/tree/main/tau_anchor"><b>Tau-Anchor</b></a> &nbsp;·&nbsp;
38
+ <a href="#citation"><b>Citation</b></a>
39
+ </p>
40
+
41
+ <br>
42
+
43
+ **CivicBench** tests detectors of AI-generated political video against the two attacks that matter in an election.
44
+ An attacker can change each pixel of a fake by a gray level or two, so that the detector passes it as real
45
+ (**evasion**). The same small change can make the detector flag real footage as generated (**framing**). Anyone who
46
+ wants to dismiss a real video can then point to the detector's verdict.
47
+
48
+ CivicBench measures both, on clips that can be shared, and one command runs it on your own detector.
49
+
50
+ <p align="center">
51
+ <img alt="Attacks on political video in both directions" src="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/teaser.jpg" width="100%">
52
+ </p>
53
+
54
+ ## Highlights
55
+
56
+ - **Two-sided by design.** Every detector is attacked in both directions, at the threshold where it is deployed.
57
+ At one gray level, four published detectors lose every fake they flagged and flag nearly every attacked real clip.
58
+ - **Political and shareable.** 10,800 clips of invented politicians, parties and countries, from 8 open generators
59
+ and 1,350 prompts across 9 kinds of civic scenes, next to 2,632 real clips under open licenses.
60
+ - **One command.** `civicbench evaluate` runs PGD, APGD and Square at any budget and prints the paper's table.
61
+ Every attacked clip is checked after rounding to 8 bits, as a viewer would see it.
62
+ - **Predictive.** A cheap attack of one gray level predicts how many clips keep their label at four.
63
+
64
+ ## Install
65
+
66
+ ```bash
67
+ pip install civicbench # core: numpy and torch
68
+ pip install "civicbench[data]" # + loading from Hugging Face
69
+ pip install "civicbench[fancy]" # + the rich progress display
70
+ ```
71
+
72
+ Python 3.10 or newer.
73
+
74
+ ## Quick start
75
+
76
+ ```python
77
+ import civicbench
78
+
79
+ clips = civicbench.load() # official test inputs and labels
80
+ results = civicbench.evaluate(my_detector, clips) # AUC, TPR, evasion and framing at 1, 2 and 4 gray levels
81
+ ```
82
+
83
+ Or from the command line, at any budgets:
84
+
85
+ ```bash
86
+ civicbench evaluate --detector my_lab.detector:build --budgets 1,2,4,8,16 --out results.json
87
+ ```
88
+
89
+ <p align="center">
90
+ <img alt="civicbench in a terminal" src="https://raw.githubusercontent.com/KurbanIntelligenceLab/CivicBench/b3d19f73b17d3aa552dba2639568fa5b49b24805/docs/assets/terminal.png" width="100%">
91
+ </p>
92
+
93
+ <sub>Output of <code>examples/demo_toy.py</code>: a toy detector on synthetic clips, so the numbers carry no meaning.
94
+ It runs in under a minute with no downloads.</sub>
95
+
96
+ ### Bring your own detector
97
+
98
+ A detector is any callable that maps frames to one score.
99
+
100
+ | | |
101
+ |---|---|
102
+ | **Input** | `torch.float32` tensor `[T, 3, H, W]`, RGB in [0, 1]. The official input is 11 frames of 512 x 512. |
103
+ | **Output** | One score. Higher means generated. A clip is flagged when its score reaches the threshold. |
104
+ | **Gradients** | PGD and APGD need a differentiable score. Square needs scores only. |
105
+
106
+ The [toolkit guide](https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/docs/toolkit.md) covers single-clip attacks, custom budgets, the survival prediction and every
107
+ difference from the paper's code.
108
+
109
+ ## Data
110
+
111
+ One open dataset, [kurbanintelligencelab/CivicBench](https://huggingface.co/datasets/kurbanintelligencelab/CivicBench), with no login and no form.
112
+
113
+ | | Clips | License |
114
+ |---|---|---|
115
+ | **Real** | 2,632 clips of briefings, speeches, legislatures, and protests | Each clip's own: public domain, CC BY or CC BY-SA |
116
+ | **Generated** | 10,800 clips from eight open text-to-video generators, with prompts, seeds, and settings | Research on detection, with each generator's terms (see the dataset card) |
117
+
118
+ Each real clip lists its creator, license, source page, start time and the changes we made. The official test input is
119
+ 11 frames of 512 x 512 per clip, stored as lossless PNG with checksums.
120
+
121
+ ## Leaderboard
122
+
123
+ CivicBench at the threshold that flags 5% of real clips. Evasion is the TPR under attack, higher is better. Framing is
124
+ the false-alarm rate on attacked real clips, lower is better. Budgets are in gray levels on the 0 to 255 scale.
125
+
126
+ | Detector | AUC | TPR | Evasion 1 | Evasion 2 | Evasion 4 | Framing 1 | Framing 2 | Framing 4 |
127
+ |:---|:---:|:---:|:---:|:---:|:---:|:---:|:---:|:---:|
128
+ | **Tau-Anchor** (ours) | **.941** | **.779** | **.731** | **.668** | **.585** | .124 | .141 | .222 |
129
+ | Mavali et al., R4 | .774 | .394 | .339 | .289 | .239 | .217 | .304 | .623 |
130
+ | Mavali et al., R2 | .768 | .381 | .236 | .164 | .135 | .261 | .536 | .884 |
131
+ | Mavali et al., ΔCLIP | .638 | .182 | .135 | .130 | .099 | **.046** | **.046** | **.101** |
132
+ | MV-MobileNet | .587 | .086 | .045 | .043 | .035 | .348 | .493 | .565 |
133
+ | STREAM-MV | .602 | .111 | .018 | .013 | .011 | .768 | .783 | .812 |
134
+ | Shahjahan et al. | .709 | .233 | .023 | .000 | .000 | .913 | .971 | .986 |
135
+ | CLIP probe | .781 | .364 | .000 | .000 | .000 | 1.000 | 1.000 | 1.000 |
136
+ | STREAM | .714 | .309 | .000 | .000 | .000 | 1.000 | 1.000 | 1.000 |
137
+ | D3 | .475 | .028 | .000 | .000 | .000 | 1.000 | 1.000 | 1.000 |
138
+ | ReStraV | .448 | .009 | .000 | .000 | .000 | .986 | 1.000 | 1.000 |
139
+
140
+ Sorted by TPR under evasion at 2 gray levels, then at 1, then by AUC. The paper adds confidence intervals, two general benchmarks and
141
+ external political footage. To add a detector, open a pull request with the `results.json` that
142
+ `civicbench evaluate --out` writes.
143
+
144
+ ## Tau-Anchor
145
+
146
+ Tau-Anchor is the detector from the paper. It fine-tunes an image encoder that was pre-trained on attacked images,
147
+ attacks real and generated clips during training, and adds a margin term for the fakes nearest its deployed threshold.
148
+ Its code, exact training recipe and model card are in [`tau_anchor/`](https://github.com/KurbanIntelligenceLab/CivicBench/tree/main/tau_anchor). The weights are on
149
+ Hugging Face at [`kurbanintelligencelab/Tau-Anchor`](https://huggingface.co/kurbanintelligencelab/Tau-Anchor).
150
+
151
+ ## Repository layout
152
+
153
+ ```
154
+ src/civicbench/ the pip package: load, attack, evaluate, predict_survival, CLI
155
+ tests/ unit tests and a golden test against the paper's attack code
156
+ examples/ demo_toy.py, a one-minute end-to-end run with no downloads
157
+ tau_anchor/ Tau-Anchor training and inference
158
+ docs/ toolkit guide and assets
159
+ ```
160
+
161
+ ## Responsible use
162
+
163
+ The release contains no attacked real clips, and thus no ready-made forgeries of real footage. The generated clips show invented people. If a clip resembles a real person, report it
164
+ through the contact on the dataset card and we will remove it.
165
+
166
+ ## Citation
167
+
168
+ ```bibtex
169
+ @misc{cakiroglu2026civicbench,
170
+ title = {CivicBench: Robust Detection of AI-Generated Political Video},
171
+ author = {Cakiroglu, Mert Onur and Buxton, Elham and Dalkilic, Mehmet and Kurban, Hasan},
172
+ year = {2026}
173
+ }
174
+ ```
175
+
176
+ ## License
177
+
178
+ The code is released under the [Apache License 2.0](https://github.com/KurbanIntelligenceLab/CivicBench/blob/main/LICENSE). The data follow the licenses in the table above and on
179
+ each dataset card. Generated clips also follow the terms of the generator that rendered them.
180
+
181
+ ## Acknowledgements
182
+
183
+ We thank the creators of the openly licensed footage, credited clip by clip in the dataset, and the developers of
184
+ CogVideoX1.5, HunyuanVideo-1.5, Kandinsky 5, LTX-Video, Mochi, SANA-Video, Wan2.1 and Wan2.2.