yrl
Datasets
All datasets matching “yrl”lvr-data-mvc_iptlvr-data-pet_ipt-p24
lvr-data-pet_ipt-p24
The IPT PET training set (weikaih/imaginative-perception-token-pet-ipt, 20,531 rows) converted to the
interleaved training schema with patch P24 of github.com/yangruoliu/mmcot-spatial: the question text now
carries its answer options ("\n(A) Closer\n(B) Further"), which the released converter dropped. Built on
2026-09-21 from commit 3e5bdce; 11 chunks x 2048 rows, 64-row row groups, parquet_info.json included.
physjepa-motion-caches
PhysJEPA motion caches
Derived caches for two-stage physics-consistent video generation on Physion.
stage1_flow_targets/ -- analytic ground-truth optical flow reconstructed from
the simulator's mesh vertices, quaternion poses and camera matrices by z-buffer
rasterisation (Drape uses flex particles + Delaunay). Not RAFT. Each entry holds
flow (7,2,160,160) in pixels, forward and cumulative from the first frame at
0.2s..1.4s, plus visibility, object_mask, future_mask, instance_id… See the full description on the dataset page: https://huggingface.co/datasets/yrlyrl/physjepa-motion-caches.lvr-data-mvc_no_thoughtlvr-data-mvc_answeronlylvr-xfer-evalcache