routing-each-image-one-several-domain-specialist-vision-backbones-trained CLM-0632observationsingle paperReviewed by AI
When routing each image to one of several domain-specialist vision backbones trained on a shared label space, a prompted MLLM router (Gemma-4-12B) reaches nearly the accuracy of a trained neural-network router and beats majority-vote fusion, because routing only needs coarse domain identification, which a general multimodal model already does zero-shot.
Capability: Getting more from several models than from one
Observed on
Four visually disjoint datasets (CIFAR10, FER2013, EuroSAT, OrganAMNIST) with a unified 38-class head; domains differ sharply enough that a ResNet-18 router hits 99.5% domain accur.
Sources
- Validated by: Unified-dataset top-1 accuracy: ARMDIL 90.78% vs best single backbone (balanced DINOv3) 89.61%, majority vote 90.47%, trained NN-Router 91.18%, oracle router 91.04%; per-domain routing accuracy 78.2-99.8%.. Measured on one four-dataset unified benchmark with one MLLM (quantized Gemma-4-12B). Margins are small (0.31 points over majority vote, -0.40 behind the NN-router) with no confidence intervals or repeated runs, and the whole ensemble sits within ~1.2 points of a single balanced DINOv3. [truncated]
Either way, bring evidence: a link, or what you saw.
Disagreeing is the most useful thing you can do here. Nothing on this page waits for a reviewer: the counts above move when evidence lands, and every source stays attached with who brought it.
Related claims
- For multi-task procedural video understanding with a small (1B) video-LLM decoder, replacing each decoder feed-forward block with task-specific expert copies and routing each video-prompt sample by its known task identity gives higher average top-1 accuracy across the five COIN tasks than the same expert topology run densely or with learned token-level or sample-level routing, with the largest gains on the procedure-forecasting tasks.Reasoning about time in video · reviewed by AI
- For deepfake video detection with 7-8B open multimodal backbones, splitting the analysis into four separately prompted perspective agents (texture, lighting, motion, physics) whose reports a judge agent reconciles beats the same backbone doing chain-of-thought or multi-turn self-questioning over the same four perspectives, and the gain is not explained by extra turns or repeated frame access.Getting more from several models than from one · reviewed by AI
- For multimodal LLMs, a behavioral perturbation signature (answer/confidence change under visual degradation, blank-image replacement, and grounding/relation probes) predicts which failure family an error belongs to far better than scalar confidence, yet folding the same signature into confidence as a direct penalty makes answer-versus-abstain ranking worse than scalar confidence alone.Saying it does not know · reviewed by AI
- On month-level egocentric recordings (20-120 days per participant) with human-written multiple-choice questions, current multimodal models degrade as the question moves from inferring recurring habits, to retrieving one specific episode, to composing evidence across days — the best model reached 71.8% macro-average against a 94.2% human baseline, with several models near or below the 25% chance level on route reasoning, cross-view spatial reasoning and direction judgement.Reasoning about time in video · reviewed by AI
- On multimodal cultural-interpretation multiple-choice questions, pipelines that pool several same-backbone evidence streams (perception, retrieval, web verification) and weight them by scored relevance beat majority voting, self-consistency and agent debate mainly on the items where the streams disagree, and the margin shrinks sharply as the backbone gets stronger.Getting more from several models than from one · reviewed by AI
Notes
Drafted from the paper by a model and filed unreviewed. Visible here so it can be read, not because anyone has vouched for it: it does not move any technique's standing and does not count toward the internal scorecard. Drafted confidence: low. Falsifier as drafted: On domain sets where visual categories are less separable, or with more experts, a prompted MLLM router falls well short of a trained router and of majority-vote fusion.