You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
Commit 8a68be8
Browse filesBrowse the repository at this point in the historyBrowse files
This section explains how to generate videos using the provided benchmark and save them in the required format. Follow the instructions below based on your model type:
77
77
78
-
#### 2.1 Image-to-Video (i2v) Models
78
+
#### 2.1 Image-to-Video Models (I2V)
79
+
80
+
<details>
81
+
<summary>I2V steps</summary>
79
82
80
83
1.**Input Requirements**:
81
84
-**Initial Frame**: Use frames from `physics-iq-benchmark/switch-frames`.
@@ -89,9 +92,12 @@ This section explains how to generate videos using the provided benchmark and sa
89
92
```
90
93
- Refer to the `generated_video_name` column in `descriptions.csv` for file naming conventions.
91
94
92
-
---
95
+
</details>
96
+
97
+
#### 2.2 Multiframe-to-Video Models (V2V)
93
98
94
-
#### 2.2 Multiframe-to-Video Models
99
+
<details>
100
+
<summary>V2V steps</summary>
95
101
96
102
1. **Input Requirements**:
97
103
- **Conditioning Frames**:
@@ -108,38 +114,29 @@ This section explains how to generate videos using the provided benchmark and sa
108
114
```
109
115
- Refer to the `generated_video_name` column in `descriptions.csv` for file naming conventions.
110
116
111
-
---
117
+
</details>
112
118
113
119
#### 2.3 Trim Generated Videos to 5 Seconds
114
120
115
-
⚠️ **IMPORTANT**: Before running the evaluation, you must trim all generated videos to **exactly 5 seconds**. Videos of any other duration will cause runtime errors.
121
+
⚠️ **IMPORTANT**: Before running the evaluation, you must trim all generated videos to **exactly 5 seconds**. Videos of any other duration are incompatible with the benchmark. If you're running V2V (=multiframe-to-video), please make sure you're not including the 3s conditioning video, only the model-generated 5 seconds.
122
+
123
+
**Example command to trip your videos using ffmpeg:**
116
124
117
-
**Trim your videos using ffmpeg:**
125
+
This example is based on cropping to the first 5 seconds which is useful for I2V; for V2V please adapt this command if the generated videos include the 3s conditioning part.
118
126
```bash
119
127
# Create output directory for trimmed videos
120
128
mkdir -p generated_videos_5s
121
129
122
-
# Trim all videos to 5 seconds at 8 FPS
123
-
# Adjust the `-r 8` parameter to match your desired FPS (8, 16, 24, or 30)
124
-
130
+
# Trim all videos to 5 seconds at desired frame rate
131
+
# Adjust the `-r 24` parameter to match your desired FPS (e.g. 8, 16, 24, or 30)
125
132
for v in generated_video_path/*.mp4; do
126
133
ffmpeg -y -i "$v" \
127
134
-t 5 \
128
-
-r 8 \
129
-
-c:v libx264 \
130
-
-pix_fmt yuv420p \
135
+
-r 24 \
131
136
"generated_videos_5s/$(basename "$v")"
132
137
done
133
138
```
134
139
135
-
**Verify video duration, FPS, and frame count using ffprobe:**
0 commit comments