solarwm-data/NOTICE at main · Enderfga/solarwm-data · GitHub

1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
SolarWM Data Engine
Copyright 2026 Guian Fang

Licensed under the Apache License, Version 2.0 (see LICENSE).

--------------------------------------------------------------------------------
Third-party source included here
--------------------------------------------------------------------------------

  vipe_patches/            Modifications to VIPE (NVIDIA, Apache-2.0), applied to a
                           VIPE checkout at setup time. These files quote fragments
                           of VIPE source as patch anchors; those fragments remain
                           under VIPE's licence and copyright.

Everything else — `solar_wm_data/`, `scripts/`, `configs/` and `skills/` — is original
to this repository.

--------------------------------------------------------------------------------
External tools this engine drives
--------------------------------------------------------------------------------

Expected under `third_party/` (see `configs/models.yaml` and `scripts/setup_*.sh`).
Not distributed here. Each keeps its own licence, which governs your use of it and
of any weights it ships — check them before redistributing anything you produce.

  VIPE            camera / SLAM backend, NVIDIA                    Apache-2.0
  Pi3X            scale-ambiguous scene structure                  weights CC BY-NC 4.0
  MoGe-2          per-frame metric depth
  DOVER           video technical + aesthetic quality              S-Lab, NON-COMMERCIAL
  UniMatch/GMFlow dense correspondence (optical flow)
  PySceneDetect   shot-cut detection                               (installed via pip)
  Qwen2.5-VL      optional local captioner                         Apache-2.0
  FFmpeg / VMAF   decoding, encoding, VMAF Motion                  the build fetched by
                  `scripts/setup_filter_tools.sh` is GPL-licensed

--------------------------------------------------------------------------------
Models used to produce the released corpus
--------------------------------------------------------------------------------

Not distributed here, and each under its own terms. Reproducing the released
annotations requires accepting them.

  Kimi-K2.6                       dense captions and the semantic annotation
                                  fields, for every clip in the corpus
  LTX-2.3                         Clean Plate generation backbone
  LTX-2.3-22b-IC-LoRA-Clean-Plate the Clean Plate adapter
  Gemma-3-12b                     text encoder in the Clean Plate pipeline,
                                  under the Gemma Terms of Use

The Clean Plate runner additionally imports the `ltx_core` / `ltx_pipelines`
packages, pinned by revision in `solar_wm_data/clean_plate/__init__.py`. They are
not installed by any script here.

--------------------------------------------------------------------------------
Datasets
--------------------------------------------------------------------------------

The source datasets this engine annotates (ABot-World-Explorer, DL3DV, MiraData,
MIND, MultiCamVideo, OmniWorld, RealCam-Vid, Sekai, SpatialVID and the other
extension sources) are NOT distributed here. Each is obtained from its own
publisher under its own terms, several of which are non-commercial or require
accepting a gated licence. This repository ships only the code that annotates them.