alpamayo
Datasets
All datasets matching “alpamayo”Alpamayo-R1-tinyDataset with 10000 samples for Alpamayo R1.
train - 80% (8000 samples)
val - 20% (2000 samples)
Each .parquet file contains 500 samples in .npy list. Get the data using the keys:
{ "uuid": uuid, "image_frames": data["image_frames"].to(torch.uint8).numpy().tobytes(), "frames_shape": list(frames.shape), "ego_history_xyz": data["ego_history_xyz"], "ego_history_rot": data["ego_history_rot"], "ego_future_xyz": data["ego_future_xyz"], "ego_future_rot": data["ego_future_rot"] }
data =… See the full description on the dataset page: https://huggingface.co/datasets/GreatGutsy/Alpamayo-R1-tiny.Alpamayo-tinyDataset with 2500 samples for Alpamayo R1. The architecture of the dataset:
/train/shard_000xx.tar
/val/shard_000yy.tar
train - 80% (2000 samples)
val - 20% (500 samples)
Each .tar file contains 25 samples in .npy.
Get the data from .npy files by using the keys:
{
"uuid": uuid,
"video_quality": "320x576",
"tokenized_data": inputs,
"ego_history_xyz": data["ego_history_xyz"],
"ego_history_rot":… See the full description on the dataset page: https://huggingface.co/datasets/GreatGutsy/Alpamayo-tiny.lead-alpamayo-curated-pai
LEAD Alpamayo-rig curated PAI
CARLA 0.9.16 expert driving from LEAD,
re-collected with the four NVIDIA Alpamayo 1.5 input cameras added, and converted to the
NVIDIA PhysicalAI-AV (PAI) on-disk format. It is read unchanged by
alpamayo-recipes Alpamayo 1.5 nav SFT
(PAIDatasetWithNav). It was built with lead2alpa
(lead2pai --curate).
train
val
routes / clips
72 / 229
6 / 24
nav samples
520
115
left / right / straight
136 / 136 / 248
31 / 31 / 53
stationary samples… See the full description on the dataset page: https://huggingface.co/datasets/earth37815/lead-alpamayo-curated-pai.mcity-av-alpamayo
mcity-av-alpamayo
Mcity AV recordings in Alpamayo-1.5 / PhysicalAI-AV format
18 × 20 s driving clips recorded at the Mcity Test Facility (Ann Arbor, MI),
converted into the layout used by nvidia/PhysicalAI-Autonomous-Vehicles, so they load
with the physical_ai_av devkit.
Clips
18 × 20 s (360 s total)
Camera
camera_front_wide_120fov, 1920×1080 H.264, 30 fps, 600 frames/clip
Egomotion
100 Hz, anchor frame, −1 s to +120 s per clip
Splits
14 train /… See the full description on the dataset page: https://huggingface.co/datasets/mcity-ai/mcity-av-alpamayo.tanitad-alpamayo2-augmentation
TanitAD Alpamayo-2-Super augmentation of PhysicalAI-AV
4,800 clips (~26.7 h of driving), 23,999 inference rows, 5 tasks per clip from
nvidia/Alpamayo2-Super run over a stratified selection of the NVIDIA
PhysicalAI-Autonomous-Vehicles dataset. NO raw sensor data is included: every row
links to the original clip by clip_id + t0_us (loader-default 5,100,000), so the
outputs can be joined back to camera/egomotion/label streams at any time.
Tasks (per clip, all outputs… See the full description on the dataset page: https://huggingface.co/datasets/Sayood/tanitad-alpamayo2-augmentation.alpamayo