LunaDream74 commited on
Commit
796a136
·
verified ·
1 Parent(s): 387dd1d

docs: dataset card with nuScenes attribution

Browse files
Files changed (1) hide show
  1. README.md +79 -0
README.md ADDED
@@ -0,0 +1,79 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ license: cc-by-nc-sa-4.0
3
+ task_categories:
4
+ - text-to-image
5
+ - image-to-text
6
+ language:
7
+ - en
8
+ tags:
9
+ - autonomous-driving
10
+ - nuscenes
11
+ - thumbnails
12
+ pretty_name: SceneSearch frame thumbnails
13
+ size_categories:
14
+ - 1K<n<10K
15
+ ---
16
+
17
+ # SceneSearch frame thumbnails
18
+
19
+ 6,858 preview thumbnails (320×180 JPEG, ~83 MB total) of `CAM_FRONT` keyframes from **170 nuScenes
20
+ v1.0-trainval scenes**. They exist to be displayed next to search results in
21
+ [SceneSearch](https://github.com/AI20K-Build-Phase-Cohort-3/P-057), a natural-language search
22
+ engine over driving footage.
23
+
24
+ **This is a cache, not a source.** The authoritative record of a frame is its coordinate —
25
+ `video_id` plus `timestamp_s` in the application's `frames.jsonl` — and every thumbnail here is
26
+ regenerable from nuScenes. Nothing is derivable *only* from this repo, which is why it lives outside
27
+ the application's git history rather than inside it.
28
+
29
+ ## Contents
30
+
31
+ One file per frame, named `<frame_id>_CAM_FRONT.jpg`, where `frame_id` is the nuScenes
32
+ `sample_data` token. Flat layout, no subdirectories.
33
+
34
+ | | |
35
+ |---|---|
36
+ | files | 6,858 |
37
+ | resolution | 320×180 (downscaled from 1600×900) |
38
+ | total size | ~83 MB |
39
+ | source scenes | 170, from `v1.0-trainval` blobs 9 and 10 |
40
+ | camera | `CAM_FRONT` only |
41
+
42
+ The 170 scenes were not chosen at random. They were selected so that rain and night appear
43
+ *independently* rather than together — in the 10-scene subset this project started from, every rainy
44
+ scene was also a night scene, which makes the two conditions impossible to tell apart when
45
+ measuring. See the project's `eval/results/report.md` for the resulting measurements.
46
+
47
+ ## Usage
48
+
49
+ ```python
50
+ from huggingface_hub import snapshot_download
51
+
52
+ snapshot_download(
53
+ repo_id="P-057-RisingStar/scenesearch-thumbs",
54
+ repo_type="dataset",
55
+ local_dir="data/thumbs",
56
+ allow_patterns=["*.jpg"],
57
+ )
58
+ ```
59
+
60
+ SceneSearch does this at **image build time**, not per request, so the running service serves
61
+ thumbnails from its own disk and never depends on this repo being reachable.
62
+
63
+ ## Licence and attribution
64
+
65
+ Derived from [**nuScenes**](https://www.nuscenes.org) by Motional, released under
66
+ [**CC BY-NC-SA 4.0**](https://creativecommons.org/licenses/by-nc-sa/4.0/). These thumbnails are a
67
+ derivative work and carry the same licence: **attribution, non-commercial, share-alike**.
68
+
69
+ > nuScenes: A multimodal dataset for autonomous driving. Holger Caesar, Varun Bankiti, Alex H. Lang,
70
+ > Sourabh Vora, Venice Erin Liong, Qiang Xu, Anush Krishnan, Yu Pan, Giancarlo Baldan, Oscar Beijbom.
71
+ > CVPR 2020.
72
+
73
+ Non-commercial use only, per the upstream licence. If you need these images commercially, go to
74
+ nuScenes directly — this repo cannot grant terms it does not hold.
75
+
76
+ ## Provenance
77
+
78
+ Produced by `scripts/publish_thumbs.py` in the SceneSearch repository. Regenerating them from a
79
+ nuScenes install and re-running that script reproduces this repo exactly.