Datasets:
ntu file removed and readme udpated
Browse files- README.md +180 -0
- json/fine_grained_pose.json +0 -0
- video/ntu.zip +0 -3
README.md
ADDED
|
@@ -0,0 +1,180 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
---
|
| 2 |
+
license: mit
|
| 3 |
+
extra_gated_prompt: >-
|
| 4 |
+
You agree to not use the dataset to conduct experiments that cause harm to
|
| 5 |
+
human subjects. Please note that the data in this dataset may be subject to
|
| 6 |
+
other agreements. Before using the data, be sure to read the relevant
|
| 7 |
+
agreements carefully to ensure compliant use. Video copyrights belong to the
|
| 8 |
+
original video creators or platforms and are for academic research use only.
|
| 9 |
+
task_categories:
|
| 10 |
+
- visual-question-answering
|
| 11 |
+
extra_gated_fields:
|
| 12 |
+
Name: text
|
| 13 |
+
Company/Organization: text
|
| 14 |
+
Country: text
|
| 15 |
+
E-Mail: text
|
| 16 |
+
modalities:
|
| 17 |
+
- Video
|
| 18 |
+
- Text
|
| 19 |
+
configs:
|
| 20 |
+
- config_name: action_sequence
|
| 21 |
+
data_files: json/action_sequence.json
|
| 22 |
+
- config_name: moving_count
|
| 23 |
+
data_files: json/moving_count.json
|
| 24 |
+
- config_name: action_prediction
|
| 25 |
+
data_files: json/action_prediction.json
|
| 26 |
+
- config_name: episodic_reasoning
|
| 27 |
+
data_files: json/episodic_reasoning.json
|
| 28 |
+
- config_name: action_antonym
|
| 29 |
+
data_files: json/action_antonym.json
|
| 30 |
+
- config_name: action_count
|
| 31 |
+
data_files: json/action_count.json
|
| 32 |
+
- config_name: scene_transition
|
| 33 |
+
data_files: json/scene_transition.json
|
| 34 |
+
- config_name: object_shuffle
|
| 35 |
+
data_files: json/object_shuffle.json
|
| 36 |
+
- config_name: object_existence
|
| 37 |
+
data_files: json/object_existence.json
|
| 38 |
+
- config_name: unexpected_action
|
| 39 |
+
data_files: json/unexpected_action.json
|
| 40 |
+
- config_name: moving_direction
|
| 41 |
+
data_files: json/moving_direction.json
|
| 42 |
+
- config_name: state_change
|
| 43 |
+
data_files: json/state_change.json
|
| 44 |
+
- config_name: object_interaction
|
| 45 |
+
data_files: json/object_interaction.json
|
| 46 |
+
- config_name: character_order
|
| 47 |
+
data_files: json/character_order.json
|
| 48 |
+
- config_name: action_localization
|
| 49 |
+
data_files: json/action_localization.json
|
| 50 |
+
- config_name: counterfactual_inference
|
| 51 |
+
data_files: json/counterfactual_inference.json
|
| 52 |
+
- config_name: fine_grained_action
|
| 53 |
+
data_files: json/fine_grained_action.json
|
| 54 |
+
- config_name: moving_attribute
|
| 55 |
+
data_files: json/moving_attribute.json
|
| 56 |
+
- config_name: egocentric_navigation
|
| 57 |
+
data_files: json/egocentric_navigation.json
|
| 58 |
+
language:
|
| 59 |
+
- en
|
| 60 |
+
size_categories:
|
| 61 |
+
- 1K<n<10K
|
| 62 |
+
---
|
| 63 |
+
# MVTamperBench Dataset
|
| 64 |
+
|
| 65 |
+
## Overview
|
| 66 |
+
|
| 67 |
+
**MVTamperBenchStart** is a robust benchmark designed to evaluate Vision-Language Models (VLMs) against adversarial video tampering effects. It leverages the diverse and well-structured MVBench dataset, systematically augmented with four distinct tampering techniques:
|
| 68 |
+
|
| 69 |
+
1. **Masking**: Overlays a black rectangle on a 1-second segment, simulating visual data loss.
|
| 70 |
+
2. **Repetition**: Repeats a 1-second segment, introducing temporal redundancy.
|
| 71 |
+
3. **Rotation**: Rotates a 1-second segment by 180 degrees, introducing spatial distortion.
|
| 72 |
+
4. **Substitution**: Replaces a 1-second segment with a random clip from another video, disrupting the temporal and contextual flow.
|
| 73 |
+
|
| 74 |
+
The tampering effects are applied to the middle of each video to ensure consistent evaluation across models.
|
| 75 |
+
|
| 76 |
+
---
|
| 77 |
+
|
| 78 |
+
## Dataset Details
|
| 79 |
+
|
| 80 |
+
The MVTamperBenchStart dataset is built upon the **MVBench dataset**, a widely recognized collection used in video-language evaluation. It features a broad spectrum of content to ensure robust model evaluation, including:
|
| 81 |
+
|
| 82 |
+
- **Content Diversity**: Spanning a variety of objects, activities, and settings.
|
| 83 |
+
- **Temporal Dynamics**: Videos with temporal dependencies for coherence testing.
|
| 84 |
+
- **Benchmark Utility**: Recognized datasets enabling comparisons with prior work.
|
| 85 |
+
|
| 86 |
+
### Incorporated Datasets
|
| 87 |
+
|
| 88 |
+
The MVTamperBenchStart dataset integrates videos from several sources, each contributing unique characteristics:
|
| 89 |
+
|
| 90 |
+
| Dataset Name | Primary Scene Type and Unique Characteristics |
|
| 91 |
+
|----------------------|-------------------------------------------------------------------------|
|
| 92 |
+
| STAR | Indoor actions and object interactions |
|
| 93 |
+
| PAXION | Real-world scenes with nuanced actions |
|
| 94 |
+
| Moments in Time (MiT) V1 | Indoor/outdoor scenes across varied contexts |
|
| 95 |
+
| FunQA | Humor-focused, creative, real-world events |
|
| 96 |
+
| CLEVRER | Simulated scenes for object movement and reasoning |
|
| 97 |
+
| Perception Test | First/third-person views for object tracking |
|
| 98 |
+
| Charades-STA | Indoor human actions and interactions |
|
| 99 |
+
| MoVQA | Diverse scenes for scene transition comprehension |
|
| 100 |
+
| VLN-CE | Indoor navigation from agent perspective |
|
| 101 |
+
| TVQA | TV show scenes for episodic reasoning |
|
| 102 |
+
|
| 103 |
+
### Dataset Expansion
|
| 104 |
+
|
| 105 |
+
The original MVBench dataset contains 3,487 videos, which have been systematically expanded through tampering effects, resulting in a total(original+tampered) of **17,435 videos**. This ensures:
|
| 106 |
+
|
| 107 |
+
- **Diversity**: Varied adversarial challenges for robust evaluation.
|
| 108 |
+
- **Volume**: Sufficient data for training and testing.
|
| 109 |
+
|
| 110 |
+
Below is a visual representation of the tampered video length distribution:
|
| 111 |
+
|
| 112 |
+

|
| 113 |
+
|
| 114 |
+
---
|
| 115 |
+
|
| 116 |
+
## Benchmark Construction
|
| 117 |
+
|
| 118 |
+
MVTamperBench is built with modularity, scalability, and reproducibility at its core:
|
| 119 |
+
|
| 120 |
+
- **Modularity**: Each tampering effect is implemented as a reusable class, allowing for easy adaptation.
|
| 121 |
+
- **Scalability**: Supports customizable tampering parameters, such as location and duration.
|
| 122 |
+
- **Integration**: Fully compatible with VLMEvalKit, enabling seamless evaluations of tampering robustness alongside general VLM capabilities.
|
| 123 |
+
|
| 124 |
+
By maintaining consistent tampering duration (1 second) and location (center of the video), MVTamperBench ensures fair and comparable evaluations across models.
|
| 125 |
+
|
| 126 |
+
---
|
| 127 |
+
|
| 128 |
+
## Download Dataset
|
| 129 |
+
|
| 130 |
+
You can access the MVTamperBenchStart dataset directly from the Hugging Face repository:
|
| 131 |
+
|
| 132 |
+
[Download MVTamperBenchStart Dataset](https://huggingface.co/datasets/Srikant86/MVTamperBenchStart)
|
| 133 |
+
|
| 134 |
+
---
|
| 135 |
+
|
| 136 |
+
## How to Use
|
| 137 |
+
|
| 138 |
+
1. Clone the Hugging Face repository:
|
| 139 |
+
```bash
|
| 140 |
+
git clone [https://huggingface.co/datasets/mvtamperbenchstart](https://huggingface.co/datasets/Srikant86/MVTamperBenchStart)
|
| 141 |
+
cd mvtamperbench
|
| 142 |
+
```
|
| 143 |
+
|
| 144 |
+
2. Load the dataset using the Hugging Face `datasets` library:
|
| 145 |
+
```python
|
| 146 |
+
from datasets import load_dataset
|
| 147 |
+
|
| 148 |
+
dataset = load_dataset("mvtamperbenchstart")
|
| 149 |
+
```
|
| 150 |
+
|
| 151 |
+
3. Explore the dataset structure and metadata:
|
| 152 |
+
```python
|
| 153 |
+
print(dataset["train"])
|
| 154 |
+
```
|
| 155 |
+
|
| 156 |
+
4. Utilize the dataset for tampering detection tasks, model evaluation, and more.
|
| 157 |
+
|
| 158 |
+
---
|
| 159 |
+
|
| 160 |
+
## Citation
|
| 161 |
+
|
| 162 |
+
If you use MVTamperBench in your research, please cite:
|
| 163 |
+
|
| 164 |
+
```bibtex
|
| 165 |
+
@misc{agarwal2024mvtamperbenchevaluatingrobustnessvisionlanguage,
|
| 166 |
+
title={MVTamperBench: Evaluating Robustness of Vision-Language Models},
|
| 167 |
+
author={Amit Agarwal and Srikant Panda and Angeline Charles and Bhargava Kumar and Hitesh Patel and Priyanranjan Pattnayak and Taki Hasan Rafi and Tejaswini Kumar and Dong-Kyu Chae},
|
| 168 |
+
year={2024},
|
| 169 |
+
eprint={2412.19794},
|
| 170 |
+
archivePrefix={arXiv},
|
| 171 |
+
primaryClass={cs.CV},
|
| 172 |
+
url={https://arxiv.org/abs/2412.19794},
|
| 173 |
+
}
|
| 174 |
+
```
|
| 175 |
+
|
| 176 |
+
---
|
| 177 |
+
|
| 178 |
+
## License
|
| 179 |
+
|
| 180 |
+
MVTamperBenchStart is built upon MVBench and therefore operates under the same license as the original MVBench. For more details, please refer to the [MVBench README](https://huggingface.co/datasets/OpenGVLab/MVBench/blob/main/README.md).
|
json/fine_grained_pose.json
DELETED
|
The diff for this file is too large to render.
See raw diff
|
|
|
video/ntu.zip
DELETED
|
@@ -1,3 +0,0 @@
|
|
| 1 |
-
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:abb90704d61716384ace900e79c24e674130548f3d25608bf12ab8075393226a
|
| 3 |
-
size 2877203075
|
|
|
|
|
|
|
|
|
|
|
|