赵智轩 commited on
Commit ·
3dd0dfd
1
Parent(s): 8bb5680
Align HF README workflow with GitHub instructions
Browse files
README.md
CHANGED
|
@@ -33,7 +33,7 @@ PerceptionComp is a benchmark for complex perception-centric video reasoning. It
|
|
| 33 |
|
| 34 |
### Dataset Description
|
| 35 |
|
| 36 |
-
PerceptionComp contains 1,114 manually annotated five-choice questions associated with 273
|
| 37 |
|
| 38 |
This Hugging Face dataset repository is intended to host the benchmark videos together with a viewer-friendly annotation file, `questions.json`, for Dataset Preview and Data Studio. The canonical annotation source, evaluation code, and model integration examples are maintained in the official GitHub repository:
|
| 39 |
|
|
@@ -68,6 +68,71 @@ PerceptionComp is not intended for:
|
|
| 68 |
- surveillance, identity inference, or sensitive attribute prediction
|
| 69 |
- modifying the benchmark protocol and reporting those results as directly comparable official scores
|
| 70 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 71 |
## Dataset Structure
|
| 72 |
|
| 73 |
### Data Instances
|
|
@@ -94,12 +159,17 @@ Core fields in each annotation item:
|
|
| 94 |
|
| 95 |
### Data Files
|
| 96 |
|
| 97 |
-
This
|
| 98 |
|
| 99 |
- `questions.json`: root-level annotation file used by Hugging Face Dataset Preview and Data Studio
|
|
|
|
| 100 |
- `README.md`: Hugging Face dataset card
|
| 101 |
- `LICENSE`: custom research-use terms for the benchmark materials
|
| 102 |
|
|
|
|
|
|
|
|
|
|
|
|
|
| 103 |
The official evaluation code prepares videos into the following local layout:
|
| 104 |
|
| 105 |
```text
|
|
@@ -112,14 +182,16 @@ Use the official download script from the GitHub repository:
|
|
| 112 |
git clone https://github.com/hrinnnn/PerceptionComp.git
|
| 113 |
cd PerceptionComp
|
| 114 |
pip install -r requirements.txt
|
| 115 |
-
|
| 116 |
```
|
| 117 |
|
|
|
|
|
|
|
| 118 |
### Data Splits
|
| 119 |
|
| 120 |
The current public release uses one official evaluation split:
|
| 121 |
|
| 122 |
-
- `test`: 1,114 multiple-choice questions over 273
|
| 123 |
|
| 124 |
## Dataset Creation
|
| 125 |
|
|
@@ -198,8 +270,8 @@ Example evaluation workflow:
|
|
| 198 |
git clone https://github.com/hrinnnn/PerceptionComp.git
|
| 199 |
cd PerceptionComp
|
| 200 |
pip install -r requirements.txt
|
| 201 |
-
|
| 202 |
-
|
| 203 |
--model YOUR_MODEL_NAME \
|
| 204 |
--provider api \
|
| 205 |
--api-key YOUR_API_KEY \
|
|
|
|
| 33 |
|
| 34 |
### Dataset Description
|
| 35 |
|
| 36 |
+
PerceptionComp contains 1,114 manually annotated five-choice questions associated with 273 referenced video IDs. The benchmark covers seven categories: outdoor tour, shopping, sport, variety show, home tour, game, and movie.
|
| 37 |
|
| 38 |
This Hugging Face dataset repository is intended to host the benchmark videos together with a viewer-friendly annotation file, `questions.json`, for Dataset Preview and Data Studio. The canonical annotation source, evaluation code, and model integration examples are maintained in the official GitHub repository:
|
| 39 |
|
|
|
|
| 68 |
- surveillance, identity inference, or sensitive attribute prediction
|
| 69 |
- modifying the benchmark protocol and reporting those results as directly comparable official scores
|
| 70 |
|
| 71 |
+
## Evaluation Workflow
|
| 72 |
+
|
| 73 |
+
The Hugging Face repository hosts the benchmark videos and the viewer-friendly test annotations. The evaluation code lives in the GitHub repository and follows this workflow:
|
| 74 |
+
|
| 75 |
+
### Step 1. Clone the Repository
|
| 76 |
+
|
| 77 |
+
```bash
|
| 78 |
+
git clone https://github.com/hrinnnn/PerceptionComp.git
|
| 79 |
+
cd PerceptionComp
|
| 80 |
+
```
|
| 81 |
+
|
| 82 |
+
### Step 2. Install Dependencies
|
| 83 |
+
|
| 84 |
+
```bash
|
| 85 |
+
pip install -r requirements.txt
|
| 86 |
+
```
|
| 87 |
+
|
| 88 |
+
### Step 3. Download the Benchmark Videos
|
| 89 |
+
|
| 90 |
+
```bash
|
| 91 |
+
python3 scripts/download_data.py --repo-id hrinnnn/PerceptionComp
|
| 92 |
+
```
|
| 93 |
+
|
| 94 |
+
If the Hugging Face dataset requires authentication:
|
| 95 |
+
|
| 96 |
+
```bash
|
| 97 |
+
python3 scripts/download_data.py \
|
| 98 |
+
--repo-id hrinnnn/PerceptionComp \
|
| 99 |
+
--hf-token YOUR_HF_TOKEN
|
| 100 |
+
```
|
| 101 |
+
|
| 102 |
+
The download helper fetches video files from the Hugging Face `data/` directory, flattens them into `benchmark/videos/`, and validates the required `video_id` set against `benchmark/annotations/1-1114.json`.
|
| 103 |
+
|
| 104 |
+
### Step 4. Run Evaluation
|
| 105 |
+
|
| 106 |
+
OpenAI-compatible API example:
|
| 107 |
+
|
| 108 |
+
```bash
|
| 109 |
+
python3 evaluate/evaluate.py \
|
| 110 |
+
--model YOUR_MODEL_NAME \
|
| 111 |
+
--provider api \
|
| 112 |
+
--api-key YOUR_API_KEY \
|
| 113 |
+
--base-url YOUR_BASE_URL \
|
| 114 |
+
--video-dir benchmark/videos
|
| 115 |
+
```
|
| 116 |
+
|
| 117 |
+
Gemini example:
|
| 118 |
+
|
| 119 |
+
```bash
|
| 120 |
+
python3 evaluate/evaluate.py \
|
| 121 |
+
--model YOUR_GEMINI_MODEL_NAME \
|
| 122 |
+
--provider gemini \
|
| 123 |
+
--api-key YOUR_GEMINI_API_KEY \
|
| 124 |
+
--video-dir benchmark/videos
|
| 125 |
+
```
|
| 126 |
+
|
| 127 |
+
### Step 5. Check the Outputs
|
| 128 |
+
|
| 129 |
+
Evaluation outputs are written to:
|
| 130 |
+
|
| 131 |
+
```text
|
| 132 |
+
evaluate/results/Results-<model>.json
|
| 133 |
+
evaluate/results/Results-<model>.csv
|
| 134 |
+
```
|
| 135 |
+
|
| 136 |
## Dataset Structure
|
| 137 |
|
| 138 |
### Data Instances
|
|
|
|
| 159 |
|
| 160 |
### Data Files
|
| 161 |
|
| 162 |
+
This Hugging Face dataset repository contains:
|
| 163 |
|
| 164 |
- `questions.json`: root-level annotation file used by Hugging Face Dataset Preview and Data Studio
|
| 165 |
+
- `data/<video_id>.<ext>`: benchmark video files downloaded by the official helper script
|
| 166 |
- `README.md`: Hugging Face dataset card
|
| 167 |
- `LICENSE`: custom research-use terms for the benchmark materials
|
| 168 |
|
| 169 |
+
The canonical annotation file used by the evaluator remains:
|
| 170 |
+
|
| 171 |
+
- `benchmark/annotations/1-1114.json` in the GitHub repository
|
| 172 |
+
|
| 173 |
The official evaluation code prepares videos into the following local layout:
|
| 174 |
|
| 175 |
```text
|
|
|
|
| 182 |
git clone https://github.com/hrinnnn/PerceptionComp.git
|
| 183 |
cd PerceptionComp
|
| 184 |
pip install -r requirements.txt
|
| 185 |
+
python3 scripts/download_data.py --repo-id hrinnnn/PerceptionComp
|
| 186 |
```
|
| 187 |
|
| 188 |
+
If your environment provides `python` instead of `python3`, use that alias consistently for the commands below.
|
| 189 |
+
|
| 190 |
### Data Splits
|
| 191 |
|
| 192 |
The current public release uses one official evaluation split:
|
| 193 |
|
| 194 |
+
- `test`: 1,114 multiple-choice questions over 273 referenced video IDs, exposed through `questions.json`
|
| 195 |
|
| 196 |
## Dataset Creation
|
| 197 |
|
|
|
|
| 270 |
git clone https://github.com/hrinnnn/PerceptionComp.git
|
| 271 |
cd PerceptionComp
|
| 272 |
pip install -r requirements.txt
|
| 273 |
+
python3 scripts/download_data.py --repo-id hrinnnn/PerceptionComp
|
| 274 |
+
python3 evaluate/evaluate.py \
|
| 275 |
--model YOUR_MODEL_NAME \
|
| 276 |
--provider api \
|
| 277 |
--api-key YOUR_API_KEY \
|