Datasets:
Upstream sources and terms
Exact acquisition sources and revisions are recorded in protocols/downloads.json.
Datasets in this repository
| Dataset | Source | Published terms |
|---|---|---|
| TartanRGBT | Official repository at the pinned revision; project page | BSD-3-Clause-Clear (dataset card) |
| VGGSound / VGGSound-309 | Oxford VGG; license text (copy in LICENSES/) |
CC BY 4.0; the copyright of the videos remains with their original owners |
| Caltech Aerial RGBT | Dataset record; DOI 10.22002/cks6g-ps927 | CC BY-NC 4.0 |
| BatVision | Dataset DOI 10.57745/HYLZNL | CC BY-SA 4.0 |
| UTD-MHAD | UT Dallas | No license published; the page asks users to cite the dataset paper |
| UCF101 | UCF CRCV; acquisition mirror | No license published; videos are from YouTube |
| NYUv2 | Official page; acquisition mirror | No license published; the page asks users to cite the dataset paper |
| MSR-VTT | Dataset paper; acquisition mirror | No license published; videos are from YouTube |
| MAVD | Dataset site; license (copy in LICENSES/) |
RL License Agreement, non-commercial use only |
Encoders
| Encoder | Source | Terms |
|---|---|---|
| ImageBind-Huge | Source at commit 53680b02; LICENSE (copy in LICENSES/) |
CC BY-NC-SA 4.0 |
| LanguageBind | Source at commit 7070c533; checkpoint revisions in protocols/downloads.json |
MIT (copy in LICENSES/) |
Terms applied to each archive
| Archives | Terms |
|---|---|
| TartanRGBT × LanguageBind (7) | BSD-3-Clause-Clear |
| TartanRGBT × ImageBind (3) | CC BY-NC-SA 4.0 |
| VGGSound × LanguageBind (1) | CC BY 4.0 |
| VGGSound × ImageBind (2) | CC BY-NC-SA 4.0 |
| Caltech Aerial RGBT × ImageBind (1) | CC BY-NC-SA 4.0 |
| BatVision × LanguageBind (1), BatVision × ImageBind (2) | CC BY-SA 4.0 |
| UTD-MHAD × ImageBind (2) | CC BY-NC-SA 4.0 |
| UCF101 × LanguageBind (1), NYUv2 × LanguageBind (2), MSR-VTT × LanguageBind (2) | CC BY-NC 4.0 |
| MAVD × ImageBind (1), MAVD × LanguageBind (1) | MAVD license, non-commercial |
Each archive contains feature vectors computed by a frozen encoder from the listed source data, together with labels and class prototypes; it contains no images, video, audio, depth maps, thermal frames or inertial signals. Use of the underlying data remains subject to each dataset's own terms.
Not included
The four Ego4D-derived archives (retrieval/imagebind/text_imu__ego4d_gyro,
retrieval/imagebind/audio_imu__ego4d_gyro and the two classification
counterparts) are not included. Ego4D is available under the
Ego4D license; licensees can
generate these archives with the code repository.