pith:L5DDZ2QZ
Perception Encoder: The best visual embeddings are not at the output of the network
The best visual embeddings for images and videos come from intermediate layers of a contrastively trained network rather than its final output.
arxiv:2504.13181 v2 · 2025-04-17 · cs.CV
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{L5DDZ2QZIV2FS2B7UNNWJLPIKJ}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
after scaling our carefully tuned image pretraining recipe and refining with our robust video data engine, we find that contrastive vision-language training alone can produce strong, general embeddings for all of these downstream tasks. There is only one caveat: these embeddings are hidden within the intermediate layers of the network.
That the intermediate-layer embeddings remain superior after the two alignment procedures without post-hoc data selection or task-specific hyperparameter tuning that would undermine the claim of a single general pretraining recipe.
Intermediate layers of a contrastively trained vision-language encoder yield stronger general embeddings than the output layer, enabling state-of-the-art performance across image/video classification, multimodal QA, and dense prediction after simple alignment.
References
Formal links
Cited by
Receipt and verification
| First computed | 2026-05-18T04:23:23.597930Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
5f463cea19457459683fa35b64ade85279a5e94f291864f9f7ba95e465291165
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/L5DDZ2QZIV2FS2B7UNNWJLPIKJ \
| 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: 5f463cea19457459683fa35b64ade85279a5e94f291864f9f7ba95e465291165
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "03ab2b855c72230580f5d0a2e514a039a1a78cf6641bf2d4636445f57968f8e2",
"cross_cats_sorted": [],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"primary_cat": "cs.CV",
"submitted_at": "2025-04-17T17:59:57Z",
"title_canon_sha256": "451081b8c383b7d3d716be07c800834b1ca1e73ae180cef305cbd11d15d32e78"
},
"schema_version": "1.0",
"source": {
"id": "2504.13181",
"kind": "arxiv",
"version": 2
}
}