Skip to content

Commit b872721

Browse files
register VideoPrism presets (keras-team#2609)
1 parent 0d3db7d commit b872721

2 files changed

Lines changed: 52 additions & 3 deletions

File tree

Lines changed: 50 additions & 1 deletion
Original file line numberDiff line numberDiff line change
@@ -1,3 +1,52 @@
11
"""VideoPrism model preset configurations."""
22

3-
backbone_presets = {}
3+
backbone_presets = {
4+
"videoprism_public_v1_base": {
5+
"metadata": {
6+
"description": (
7+
"114 million parameter, 12-layer ViT-B, 16-frame, 288x288 "
8+
"resolution, video-only encoder for "
9+
"spatiotemporal representation."
10+
),
11+
"params": 114000000,
12+
"path": "video_prism",
13+
},
14+
"kaggle_handle": "kaggle://keras/videoprism/keras/videoprism_public_v1_base/1",
15+
},
16+
"videoprism_public_v1_large": {
17+
"metadata": {
18+
"description": (
19+
"354 million parameter, 24-layer ViT-L, 16-frame, 288x288 "
20+
"resolution, video-only encoder for "
21+
"spatiotemporal representation."
22+
),
23+
"params": 354000000,
24+
"path": "video_prism",
25+
},
26+
"kaggle_handle": "kaggle://keras/videoprism/keras/videoprism_public_v1_large/1",
27+
},
28+
"videoprism_lvt_public_v1_base": {
29+
"metadata": {
30+
"description": (
31+
"248 million parameter, 12-layer ViT-B video encoder + text "
32+
"encoder, 16-frame, 288x288 resolution, for multimodal "
33+
"video-language tasks."
34+
),
35+
"params": 248000000,
36+
"path": "video_prism",
37+
},
38+
"kaggle_handle": "kaggle://keras/videoprism/keras/videoprism_lvt_public_v1_base/1",
39+
},
40+
"videoprism_lvt_public_v1_large": {
41+
"metadata": {
42+
"description": (
43+
"580 million parameter, 24-layer ViT-L video encoder + text "
44+
"encoder, 16-frame, 288x288 resolution, for multimodal "
45+
"video-language tasks."
46+
),
47+
"params": 580000000,
48+
"path": "video_prism",
49+
},
50+
"kaggle_handle": "kaggle://keras/videoprism/keras/videoprism_lvt_public_v1_large/1",
51+
},
52+
}

tools/checkpoint_conversion/convert_video_prism_checkpoints.py

Lines changed: 2 additions & 2 deletions
Original file line numberDiff line numberDiff line change
@@ -2,12 +2,12 @@
22
Convert VideoPrism checkpoints to the Keras format.
33
44
The official repo is here:
5-
https://github.com/google-research/videoprism
5+
https://github.com/google-deepmind/videoprism
66
77
Setup:
88
99
```shell
10-
git clone https://github.com/google-research/videoprism.git
10+
git clone https://github.com/google-deepmind/videoprism.git
1111
```
1212
1313
Usage:

0 commit comments

Comments
 (0)