Pith. sign in

REVIEW 15 cited by

RAFT: Recurrent All-Pairs Field Transforms for Optical Flow

Not yet reviewed by Pith; the record is open.

This paper has not been read by Pith yet. Machine review is queued; the pith claim, tier, and objections will appear here once it completes.

SPECIMEN: schema-true, not a live event

T0 review · schema-true

One-sentence machine reading of the paper's core claim.

pith:XXXXXXXX · record.json · timestamp

arxiv 2003.12039 v3 pith:LDXO7ENT submitted 2020-03-26 cs.CV

RAFT: Recurrent All-Pairs Field Transforms for Optical Flow

classification cs.CV
keywords rafterrorfieldflowpixelsrecurrentachievesall-pairs
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved
0 comments
Share X Bluesky LinkedIn Reddit HN
read the original abstract

We introduce Recurrent All-Pairs Field Transforms (RAFT), a new deep network architecture for optical flow. RAFT extracts per-pixel features, builds multi-scale 4D correlation volumes for all pairs of pixels, and iteratively updates a flow field through a recurrent unit that performs lookups on the correlation volumes. RAFT achieves state-of-the-art performance. On KITTI, RAFT achieves an F1-all error of 5.10%, a 16% error reduction from the best published result (6.10%). On Sintel (final pass), RAFT obtains an end-point-error of 2.855 pixels, a 30% error reduction from the best published result (4.098 pixels). In addition, RAFT has strong cross-dataset generalization as well as high efficiency in inference time, training speed, and parameter count. Code is available at https://github.com/princeton-vl/RAFT.

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Forward citations

Cited by 15 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score.

  1. WorldRoamBench: An Open-World Benchmark for Long-Horizon Stability of Interactive World Models

    cs.CV 2026-06 accept novelty 7.0

    A 600+ case open-world benchmark finds no interactive world model is simultaneously action-faithful, visually stable, physically plausible, and memory-consistent over long WASD interaction.

  2. WorldRoamBench: An Open-World Benchmark for Long-Horizon Stability of Interactive World Models

    cs.CV 2026-06 unverdicted novelty 7.0

    WorldRoamBench is a new benchmark for interactive world models that evaluates four stability dimensions with custom metrics and finds no tested model performs reliably across all.

  3. CaRLi-V: Camera-RADAR-LiDAR Point-Wise 3D Velocity Estimation

    cs.RO 2025-11 unverdicted novelty 7.0

    CaRLi-V fuses RADAR velocity cube, camera optical flow, and LiDAR ranges in a closed-form solution to produce dense point-wise 3D velocity estimates that outperform scene flow methods on a custom dataset.

  4. Vision-Language-Motion Maps: An Open-Vocabulary, Uncertainty-Aware, Queryable Motion Attribute for 3D Scene Maps

    cs.RO 2026-07 conditional novelty 6.0

    VLMM is a 3D map representation where each object carries a fused, uncertainty-aware motion attribute (language-based movability prior + observed geometric motion) that makes motion queries such as 'what is moving' an...

  5. WorldRoamBench: An Open-World Benchmark for Long-Horizon Stability of Interactive World Models

    cs.CV 2026-06 unverdicted novelty 6.0

    WorldOdysseyBench introduces four new evaluation dimensions and metrics for interactive world models and shows that none of 10+ tested models reliably pass all of them.

  6. Zero-Parameter Geometric Gating for Temporally Stable Low-Altitude UAV Video Semantic Segmentation

    cs.CV 2026-06 unverdicted novelty 6.0

    A RANSAC-based geometric gate routes regions to homography or optical flow warping before SSP fusion, improving mIoU by 4.24-4.91% on synthetic UAVid with only 211K added parameters to frozen backbones.

  7. Turning Video Models into Generalist Robot Policies

    cs.RO 2026-05 unverdicted novelty 6.0

    Decouples action-free video world models from embodiment-specific IDMs using Jacobian-based translation to achieve zero-shot cross-embodiment robot policies.

  8. EgoDyn-Bench: Evaluating Ego-Motion Understanding in Vision-Centric Foundation Models for Autonomous Driving

    cs.CV 2026-04 unverdicted novelty 6.0

    EgoDyn-Bench reveals a perception bottleneck in vision-centric foundation models: ego-motion logic derives from language while visual input adds negligible signal, with explicit trajectories restoring consistency.

  9. EgoDyn-Bench: Evaluating Ego-Motion Understanding in Vision-Centric Foundation Models for Autonomous Driving

    cs.CV 2026-04 unverdicted novelty 6.0

    EgoDyn-Bench finds a Perception Bottleneck: foundation models hold ego-motion logic in language but misalign it with vision, underperforming geometric baselines until given explicit trajectories.

  10. Interpreting Video Representations with Spatio-Temporal Sparse Autoencoders

    cs.CV 2026-04 conditional novelty 6.0

    Spatio-temporal contrastive SAEs recover temporal coherence lost by hard TopK, improve action probes by +3.9% and retrieval by up to 2.8× R@1, and expose a monosemanticity metric artifact.

  11. VisPhyWorld: Probing Physical Reasoning via Code-Driven Video Reconstruction

    cs.CV 2026-02 unverdicted novelty 6.0

    VisPhyWorld evaluates MLLMs' physical reasoning via executable code generation for video reconstruction, with VisPhyBench showing strong semantics but weak parameter inference and dynamics simulation.

  12. Learning Zero-Shot Subject-Driven Video Generation Using 1% Compute

    cs.CV 2025-04 unverdicted novelty 6.0

    A zero-shot subject-driven video generation framework that decomposes the task into identity injection from 200K subject-image pairs and motion preservation from 4K arbitrary videos, trained in 288 A100 GPU hours on C...

  13. A Survey on Evaluating Quality and Trustworthiness in LLM-Generated Data

    cs.AI 2026-01 conditional novelty 5.0

    A metric-oriented survey that classifies intrinsic quality and trustworthiness metrics for LLM-generated data across six modalities and documents systematic evaluation gaps in the current literature.

  14. AV1 Motion Vector Fidelity and Application for Efficient Optical Flow

    eess.IV 2025-10 unverdicted novelty 5.0

    AV1 motion vectors serve as a high-fidelity warm-start for the RAFT optical flow network, delivering a four-fold speedup in convergence with only minor end-point error increase compared to standard initialization.

  15. A Survey on Vision-Language-Action Models: An Action Tokenization Perspective

    cs.RO 2025-07 unverdicted novelty 5.0

    The survey frames VLA models as pipelines that generate progressively grounded action tokens and classifies those tokens into eight types to guide future development.