Pith Number

pith:K4CIF6EL

pith:2026:K4CIF6ELOWVEHIM7A4BOYW5EGV

not attested not anchored not stored refs resolved

MPM: Mutual Pair Merging for Efficient Vision Transformers

David Rousseau, Pejman Rasti, Simon Rav\'e

Mutual Pair Merging shortens vision transformer sequences for semantic segmentation by averaging mutual nearest-neighbor token pairs while preserving reconstruction for existing decoders.

arxiv:2604.05718 v1 · 2026-04-07 · cs.CV

Open paper page JSON Open Graph Bundle Merged state Verified badge What is a Pith Number?

Add to your LaTeX paper

\usepackage{pith}
\pithnumber{K4CIF6ELOWVEHIM7A4BOYW5EGV}

Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge

Record completeness

1 Bitcoin timestamp

2 Internet Archive

3 Author claim open · sign in to claim

4 Citations open

5 Replications open

✓ Portable graph bundle live · download bundle · merged state

The bundle contains the canonical record plus signed events. A mirror can host it anywhere and recompute the same current state with the deterministic merge algorithm.

Claims

C1strongest claim

On ADE20K, MPM reduces per-image latency by up to 60% for ViT-Tiny on Raspberry Pi 5, and increases throughput by up to 20% on H100 with FlashAttention-2 while keeping the mIoU drop below 3%.

C2weakest assumption

That the overhead of computing mutual nearest-neighbor pairs and the subsequent gather-based reconstruction remains small enough on the target hardware to produce net latency gains, and that the merge map allows existing segmentation heads to be used unchanged without further accuracy degradation.

C3one line summary

MPM merges mutual nearest-neighbor token pairs in cosine space for ViTs, records a merge map for reconstruction, and delivers up to 60% latency reduction on Raspberry Pi 5 and 20% throughput gain on H100 with under 3% mIoU drop on ADE20K.

References

39 extracted · 39 resolved · 0 Pith anchors

[1] Token cropr: Faster vits for quite a few tasks 2025

[2] Token merging: Your vit but faster 2023

[3] Vision transformer adapter for dense predictions 2023

[4] Schwing, Alexan- der Kirillov, and Rohit Girdhar 2022

[5] IEEE, 2022. 1, 2, 7, 8 2022

Formal links

2 machine-checked theorem links

Receipt and verification

First computed	2026-06-03T01:05:50.228775Z
Builder	pith-number-builder-2026-05-17-v1
Signature	Pith Ed25519 (`pith-v1-2026-05`) · public key
Schema	pith-number/v1.0

Canonical hash

570482f88b75aa43a19f0702ec5ba435602d63f91baae81575c8c22a8e57762a

Aliases

arxiv: 2604.05718 · arxiv_version: 2604.05718v1 · doi: 10.48550/arxiv.2604.05718 · pith_short_12: K4CIF6ELOWVE · pith_short_16: K4CIF6ELOWVEHIM7 · pith_short_8: K4CIF6EL

Agent API

Resolver JSON Graph JSON Events JSON Schema Signing key

Verify this Pith Number yourself

curl -sH 'Accept: application/ld+json' https://pith.science/pith/K4CIF6ELOWVEHIM7A4BOYW5EGV \
  | jq -c '.canonical_record' \
  | python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: 570482f88b75aa43a19f0702ec5ba435602d63f91baae81575c8c22a8e57762a

Canonical record JSON

{
  "metadata": {
    "abstract_canon_sha256": "7968cca217d129e3bae8b2cd1b9e35f83ac114439fddf5cb883732628f028cbd",
    "cross_cats_sorted": [],
    "license": "http://creativecommons.org/licenses/by-sa/4.0/",
    "primary_cat": "cs.CV",
    "submitted_at": "2026-04-07T11:16:18Z",
    "title_canon_sha256": "9cae031688c14722c93202bc3e868fec19939cee50f2666bf964b449622204af"
  },
  "schema_version": "1.0",
  "source": {
    "id": "2604.05718",
    "kind": "arxiv",
    "version": 1
  }
}