{"id":"9d63973d-f2fd-4e11-91d1-49d2576a8cf8","arxiv_id":"2605.30561","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"high","formal_verification":"none","parameter_count":0,"one_line_summary":"Standard VLMs achieve expert-level 3D performance on depth estimation, pose estimation, and object understanding via three simple techniques without architecture changes or regression losses.","lead":"The paper claims that standard Vision Language Models can master diverse 3D tasks using only focal length unification, text-based pixel references, and data mixture and scaling. This approach could simplify 3D AI development by avoiding specialized expert models and complex training setups.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.3","headline":"Isolation of the three factors from confounders (model size, training details, data quality) is not demonstrated","rationale":"The reader's weakest_assumption already identifies the exact isolation requirement needed for the 'all you need' and 'not necessary conditions' statements; the abstract alone supplies no counter-evidence, so the concern stands and the UNVERDICTED verdict is unaffected.","tokens_in":1790,"tokens_out":293,"duration_ms":16338,"concrete_test":"Locate the ablation or scaling sections; extract any table that reports depth or correspondence metrics while varying only one of the three factors at a time with fixed model architecture, parameter count, optimizer, and data source; verify whether the reported delta remains >0.05 when those are held constant.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim requires that focal length unification, text-based pixel reference, and data mixture/scaling are jointly sufficient and that architecture changes, model scale, augmentations, and regression losses are unnecessary. This holds only if the large-scale study ablated while holding all other variables fixed. The abstract states performance gains (0.84 -> 0.9 on depth) and matching expert accuracy but provides no description of controls, ablation tables, or whether model size/training recipes were matched across conditions. Without such isolation, gains could arise from unmentioned factors rather than the three listed ones.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.3","summary":"The paper claims that Vision Language Models are native 3D learners. Its central argument, based on an in-depth large-scale study, is that focal length unification, text-based pixel reference, and data mixture/scaling are jointly sufficient for effective 3D learning in standard VLMs. It proposes the simple VLM3 method, which reportedly advances depth estimation accuracy from 0.84 to 0.9 and matches expert vision model performance on pixel correspondence, camera pose estimation, and object-level 3D tasks, while showing that architecture changes, larger models, heavy augmentations, and complex losses (including regression) are unnecessary.","tokens_in":1888,"tokens_out":367,"duration_ms":19200,"significance":"If the empirical isolation of the three factors holds under controlled conditions, the result would be significant: it would indicate that standard VLMs can handle diverse 3D tasks via minimal, text-based adaptations and data strategies, challenging the necessity of specialized 3D architectures and potentially enabling a simpler, more scalable paradigm.","major_comments":[{"comment":"Abstract: The claim that focal length unification, text-based pixel reference, and data mixture/scaling are all that is needed (and that architecture changes, model scale, augmentations, and complex losses are not necessary) rests on an asserted 'in-depth large scale study,' yet the manuscript supplies no methodology details, dataset descriptions, ablation tables, or controls that hold model size, training recipes, and other variables fixed across conditions. This is load-bearing for the central claim, as performance gains (e.g., the cited depth improvement) cannot be attributed to the three factors without such isolation.","section":"Abstract"}],"minor_comments":[],"recommendation":"major_revision","confidential_remarks":null},"author_rebuttal":{"model":"grok-4.3","summary":"We thank the referee for the thoughtful review and constructive feedback. We address the major comment below and agree that greater clarity on the experimental design is warranted to support the central claims.","responses":[{"response":"We agree that the abstract is high-level and that explicit isolation of the three factors requires clear documentation of controls. The full manuscript contains an experimental section with dataset descriptions, training details, and ablation studies; however, these may not sufficiently highlight fixed variables (model size, recipes) or directly attribute gains to focal length unification, text-based referencing, and data scaling. We will revise by expanding the methods/experiments section with a dedicated controlled-study subsection, additional ablation tables that explicitly hold other factors fixed, and clearer result attribution. This addresses the load-bearing concern without altering the core findings.","revision_made":"yes","referee_comment":"The claim that focal length unification, text-based pixel reference, and data mixture/scaling are all that is needed (and that architecture changes, model scale, augmentations, and complex losses are not necessary) rests on an asserted 'in-depth large scale study,' yet the manuscript supplies no methodology details, dataset descriptions, ablation tables, or controls that hold model size, training recipes, and other variables fixed across conditions. This is load-bearing for the central claim, as performance gains (e.g., the cited depth improvement) cannot be attributed to the three factors without such isolation."}],"tokens_in":1381,"tokens_out":312,"duration_ms":20270,"standing_objections":[]},"desk_editor":{"model":"grok-4.3","letter":"Hi,\n\nThe main thing to know is that this paper argues standard VLMs are already native 3D learners. The authors say focal length unification, text-based pixel references, and data mixture plus scaling are enough to reach expert-level performance on depth estimation and tasks like pixel correspondence, camera pose, and object-level 3D understanding. They report lifting depth accuracy from 0.84 to 0.9 while keeping the original architecture and avoiding regression losses or heavy augmentations.\n\nWhat the work does well is lay out a minimal recipe that unifies several 3D capabilities under one prompting and training approach. The simplicity is useful for anyone who wants to avoid building separate expert models for robotics or scene tasks. The claim that many standard 3D design choices are unnecessary is a direct challenge to current practice and worth testing.\n\nThe soft spots sit in the evidence. The abstract describes an in-depth large-scale study but supplies no dataset details, ablation tables, or controls for model size and training recipe. The stress-test concern is fair: without matched conditions it is hard to know whether the gains trace to the three listed factors or to other unmentioned changes. If the full paper contains clear isolation experiments, that would fix the gap; otherwise the central result stays hard to trust.\n\nThis paper is for VLM researchers who want practical ways to add 3D without new architectures. Readers looking for simple baselines or scaling recipes would find it relevant.\n\nIt deserves a serious referee because the claim is concrete and the potential payoff is large if the controls hold. I recommend sending it to review but asking the authors to add explicit ablation results and matched training details in the next version.\n\nBest,","headline":"The paper claims VLMs handle multiple 3D tasks with just focal length unification, text pixel references, and data scaling, but the abstract leaves the isolation of those factors unverified.","tokens_in":2401,"tokens_out":431,"would_cite":false,"duration_ms":26077,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.3","headline":"Vision language models learn 3D tasks using only focal length unification, text-based pixel references, and data scaling.","keywords":["vision language models","3D understanding","depth estimation","focal length unification","text-based pixel reference","data scaling","VLM3"],"falsifier":"A controlled experiment applying focal length unification, text-based pixel reference, and data mixture and scaling to a VLM that shows no improvement in depth estimation or other 3D metrics beyond the baseline.","tokens_in":2687,"feed_emoji":"📐","tokens_out":619,"duration_ms":20669,"temperature":0.7,"pith_summary":"This paper claims that vision language models are inherently suited for 3D understanding and do not require specialized designs. Through large-scale studies, it finds that unifying focal lengths across data, referencing pixels via text, and carefully mixing and scaling training data enable strong performance on 3D tasks. These simple steps allow standard VLMs to handle depth estimation, camera pose, and object 3D understanding at levels matching expert models. The approach avoids complex losses, heavy augmentations, or architecture modifications that are common in traditional 3D vision work.","feed_headline":"Three tweaks let VLMs handle 3D tasks like experts","feed_subtitle":"Focal length unification, text pixel references and data scaling enable standard models to match specialized 3D vision performance.","key_machinery":"The three enabling factors of focal length unification, text-based pixel reference, and data mixture and scaling that allow standard VLMs to master 3D tasks through text-based training.","core_discovery":"VLMs are native 3D learners where focal length unification, text-based pixel reference, and data mixture and scaling suffice for effective 3D learning, rendering model architecture changes, larger models, heavy data augmentations, and complex losses unnecessary.","pith_inferences":["This approach may allow 3D capabilities to emerge in general multimodal models without dedicated 3D training pipelines.","It could simplify integration of 3D understanding into applications like autonomous navigation by reusing existing VLM infrastructure.","Testing these factors on even larger VLMs or different data distributions might reveal further performance gains."],"forward_implications":["Depth estimation accuracy improves substantially from 0.84 to 0.9 on standard benchmarks.","Pixel correspondence, camera pose estimation, and object-level 3D understanding reach accuracy levels comparable to expert vision models.","Standard VLM architectures and text-based training suffice without task-specific modifications.","VLM3 provides a scalable method for diverse 3D tasks using the simplest design."],"fun_headline_variants":["VLMs learn 3D natively with focal unification text refs scaling","Three elements suffice for VLMs 3D learning without model changes","Focal unification and text pixel refs enable VLMs as 3D learners","Data scaling with text refs allows VLMs to match 3D expert models"],"cache_read_input_tokens":2112,"weakest_assumption_plain":"The observed performance gains on 3D tasks result solely from focal length unification, text-based pixel reference, and data mixture and scaling, isolated from other training variables.","fun_headline_variants_meta":{"raw":{"variants":["VLMs learn 3D natively with focal unification text refs scaling","Three elements suffice for VLMs 3D learning without model changes","Focal unification and text pixel refs enable VLMs as 3D learners","Data scaling with text refs allows VLMs to match 3D expert models"]},"model":"grok-4.3","cost_usd":0.00653,"raw_usage":{"total_tokens":3048,"prompt_tokens":657,"num_sources_used":0,"completion_tokens":78,"cost_in_usd_ticks":65299500,"prompt_tokens_details":{"text_tokens":657,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":2313,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":657,"tokens_out":78,"duration_ms":17338,"temperature":1.0,"reasoning_tokens":2313,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-06-29T07:43:22.756531+00:00","model_set":{"reader":"grok-4.3"},"falsifier":"A controlled experiment applying focal length unification, text-based pixel reference, and data mixture and scaling to a VLM that shows no improvement in depth estimation or other 3D metrics beyond the baseline.","supporting_citations":[],"review_version":1}