vace-object / UPSTREAM_TERMS.md
calisolo's picture
Add UPSTREAM_TERMS.md
ade3c56 verified
|
Raw History Blame Contribute Delete
1.96 kB

Upstream sources and terms

This compilation modifies source data through cropping, rectangular masking, white-background compositing, estimated HC-STVG segmentation, and caption corrections. Preserve source attribution and applicable source terms. Original authors do not endorse this compilation.

  • DAVIS 2017 — Jordi Pont-Tuset et al., The 2017 DAVIS Challenge on Video Object Segmentation. Project, CC BY-NC 4.0. The distributed DAVIS 2017 README states CC BY-NC 4.0; source-video links may carry additional terms. README and SOURCES are included where available.
  • MeViS — Henghui Ding et al., MeViS: A Large-scale Benchmark for Video Segmentation with Motion Expressions, ICCV 2023, and MeViS v2 updates. Project, official repository, CC BY-NC-SA 4.0. The official README states non-commercial research use only; derived MeViS assets retain those terms. Download: FudanCVL/MeViSv2 revision ed94977691d86cb670c47815488db6f46709a40d.
  • HC-STVG 2.0 — Official repository. Download mirrors: CircleRadon/hc-stvg video revision 4d26c230bbdba17e57a2e9965d15b584cb69a33e; ShijianW01/hc-stvg2 annotation revision a29770dd8321a64063b183b486d71fbcf95b85e4. This compilation asserts no uniform license for underlying videos; original dataset/source-video terms and rights apply.
  • Ref-YouTube-VOS — Seonguk Seo et al., URVOS: Unified Referring Video Object Segmentation Network with a Large-Scale Benchmark, ECCV 2020. Official dataset. Download mirror: blue7012/ref-youtube-vos revision 2beceace712523905138314bf0ae6958e8e4cfc8. Underlying YouTube-VOS/source-video terms apply; this compilation asserts no uniform source-video license.