Skip to content

Add curated NuRec train/val splits for local evaluation - #167

Merged
mwatson-nvidia merged 1 commit into
NVlabs:e2e_challengefrom
01Dami23:curated-nurec-splits
Aug 31, 2026
Merged

mwatson-nvidia merged 1 commit into
NVlabs:e2e_challengefrom
01Dami23:curated-nurec-splits

Conversation

@01Dami23

Copy link
Copy Markdown

Contributes an 80/20 train/validation split over the public NuRec 26.01 and 26.04 releases, so challenge participants can measure a policy locally between leaderboard submissions on scenes whose HD maps have been checked.

Scenes recorded on the same road are grouped and assigned to one suite as a unit, so a policy is not evaluated on a road it trained on.

Follows the existing per-release catalog convention (similarly to sim_scenes_2505.csv) and the documented mechanism for using a non-default catalog: pinning scenes.scenes_csv / scenes.suites_csv packaged behind a config group so the shared catalogs stay unmodified and nothing loads unless it is selected:

uv run alpasim_wizard +e2e_challenge=dev +nurec_scenes=curated_val ...

data/scenes/sim_scenes.csv and data/scenes/sim_suites.csv are untouched.

NOTE: the added 26.04 catalogue doesn't include all 1607 as some scenes were duplicates of release 26.01. To ensure the scenes from 26.01 are considered for duplicates I excluded them from the 26.04 list. If you would rather take the 26.04 version, I can easily add the full 1607 scenes and alpasim automatically selects the latest ones in case of duplicates, which would be the 26.04 ones.

Contributes an 80/20 train/validation split over the public NuRec 26.01 and
26.04 releases, so challenge participants can measure a policy locally between
leaderboard submissions on scenes whose HD maps have been checked.

2202 of 2364 scenes survive; 162 are excluded, mostly for defective maps (road
geometry missing under the vehicle, intersecting lane lines, map absent at the
start or end of the route) rather than impossible drives.

Scenes recorded on the same road are grouped and assigned to one suite as a
unit, so a policy is not evaluated on a road it trained on: 876 same-road pairs
among the 2202 scenes, 0 straddling the two suites.

Follows the existing per-release catalog convention (sim_scenes_2505.csv) and
the documented mechanism for using a non-default catalog -- pinning
scenes.scenes_csv / scenes.suites_csv -- packaged behind a config group so the
shared catalogs stay unmodified and nothing loads unless it is selected:

    uv run alpasim_wizard +e2e_challenge=dev +nurec_scenes=curated_val ...

data/scenes/sim_scenes.csv and data/scenes/sim_suites.csv are untouched.
@copy-pr-bot

copy-pr-bot Bot commented Aug 28, 2026

Copy link
Copy Markdown

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.

@mwatson-nvidia
mwatson-nvidia merged commit 85d5cb1 into NVlabs:e2e_challenge Aug 31, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants