{"id":"b6f18e9e-29d1-4c03-a374-f47822f4fbe2","arxiv_id":"2607.08891","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A VLM maps orthographic drawings to executable CadQuery, reaching 100% valid code and ~7–8% relative IoU gains over the next-best baselines on DeepCAD and Fusion 360.","lead":"Ortho2CAD turns raster multi-view engineering drawings into editable CadQuery programs that compile to 3D CAD. It pairs supervised fine-tuning where code labels exist with geometry-reward RL where they do not, and ships a drawing generator for scale.","discovery_kind":"new_application","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified beyond the reader's already-flagged synthetic-proxy and small-subset limits; the central IoU/validity claim is internally consistent on the stated benchmarks.","rationale":"The central claim is a measured empirical superiority on synthetic orthographic-to-CadQuery benchmarks, not a universal claim about all industrial drawings. The dual-regime design (SFT when GenCAD-Code labels exist; Dr. GRPO-style sequence-level RL with IoU reward when only STEP exists) is coherent, and 100% validity plus the reported IoU deltas are the quantities that must hold. The reader's weakest assumption correctly flags the synthetic, sketch-extrude-only, sparsely dimensioned, 100-sample evaluation regime as the main transfer risk; that risk is already reflected in the CONDITIONAL verdict and MODERATE confidence. I find no stronger load-bearing flaw (e.g., non-comparable baselines, non-reproducible metric definition, or reward that trivially gameable without geometry) that would force a harsher verdict. Therefore the reader's CONDITIONAL assessment stands; the concrete check above would simply confirm the numbers under independent execution.","tokens_in":13928,"tokens_out":592,"duration_ms":6264,"concrete_test":"Independently re-render the same 100 DeepCAD and 100 Fusion 360 test STEP files with the published pythonOCC pipeline, execute the authors' released Ortho2CAD checkpoints (or retrain under the stated hyperparameters), and recompute mean IoU and valid-code rate against CAD-Coder and GPT-5.2 under identical alignment; if Ortho2CAD no longer leads by ~7% relative IoU or validity drops below 100%, the headline claim fails on its own terms.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's strongest claim is empirical and narrowly scoped: on the authors' synthetic first-angle orthographic drawings (three views, dashed hidden lines, three bounding-box dimensions only) derived from DeepCAD/GenCAD-Code and Fusion 360 sketch-and-extrude STEP files, Ortho2CAD yields 100% executable CadQuery and mean solid IoU above the listed baselines (Tables 3–4; 100-example subsets). That claim is supported by the reported numbers, the dual SFT/RL recipe (§3.2–3.3), and the open generation pipeline. The reader's weakest assumption—that these drawings are a sufficient proxy for industrial raster drawings—is real and correctly identified in §3.1 and §5, but it is a scope/transfer limitation rather than an internal inconsistency that would falsify the reported gains on the synthetic benchmarks themselves. No hidden mathematical contradiction, reward-hacking proof, or evaluation-protocol error that would reverse the ranking on the stated test subsets is evident from the text.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper presents Ortho2CAD, a VLM pipeline that maps rasterized first-angle orthographic drawings (three views with dashed hidden lines and three bounding-box dimensions) to executable CadQuery code and thus editable 3D CAD. When GenCAD-Code labels exist (DeepCAD), the authors supervised-fine-tune Qwen3-VL-8B-Instruct with standard cross-entropy (Eq. 1). When only STEP geometry is available (Fusion 360 Reconstruction), they initialize from the SFT model and optimize a Dr. GRPO / GSPO-style sequence-level objective with mean-centered group advantages and solid IoU reward (Eqs. 2–6). A pythonOCC generator produces the drawings at scale. On 100-example test subsets matching prior CAD-Coder protocol, the model reports 100% valid executable code and mean IoU of 0.7922 (DeepCAD SFT) and 0.5601 (Fusion RL), exceeding the listed baselines by roughly 7–8% relative IoU.","tokens_in":14269,"tokens_out":1156,"duration_ms":11482,"significance":"If the reported gains hold, the work supplies a practical, open pathway from the dominant industrial communication format (raster orthographic drawings) to editable parametric CAD, covering both code-supervised and code-free regimes. Strengths that should be credited include: (i) a public pythonOCC drawing generator with hidden lines and key dimensions (>150k samples), (ii) explicit dual-regime training (SFT + geometry-grounded RL) with fully stated objectives and hyperparameters, (iii) 100% code validity on the evaluation subsets, and (iv) released code. These elements make the empirical claim reproducible and useful as a baseline for drawing-to-CAD research, even though the drawings remain synthetic and sketch-and-extrude-only.","major_comments":[{"comment":"§3.1 and §5 correctly note that drawings carry only three bounding-box dimensions and that sources are limited to sketch-and-extrude STEP (DeepCAD / Fusion 360). The central claim of superiority is therefore scoped to this synthetic proxy. The manuscript should state more sharply in the abstract and introduction that transfer to fully dimensioned industrial drawings, incomplete views, or richer B-rep operations is untested, so that the 7% relative IoU claim is not read as a general manufacturing result.","section":null},{"comment":"Tables 3–4 and §3.2–3.3 evaluate only the same 100-example subsets used by CAD-Coder, citing compute cost. While consistent with prior work, this leaves ranking variance and confidence intervals unreported. A major revision should either (a) report IoU mean/std over the full official test splits (or at least a substantially larger random subset) or (b) provide bootstrap/CI estimates on the 100-example sets so that the 7–8% relative gains can be assessed for statistical stability.","section":null},{"comment":"§3.3 and Table 4: the RL model is initialized from DeepCAD SFT and then trained on Fusion geometry rewards. The large jump from Ortho2CAD DeepCAD SFT (IoU 0.3697) to Ortho2CAD RL (0.5601) is the key evidence that geometry-grounded RL compensates for missing code labels. An ablation that starts RL from the untuned Qwen3-VL backbone (or reports intermediate checkpoints) is needed to separate the contribution of the SFT warm-start from the RL objective itself; without it the load-bearing claim that RL alone enables unsupervised adaptation remains only partially supported.","section":null}],"minor_comments":[{"comment":"Abstract and §1: “average relative improvement of over 7%” aggregates DeepCAD (7.6%) and Fusion (8.1%); state the two numbers separately for clarity.","section":null},{"comment":"§3.2: the fixed prompt is acknowledged as suboptimal; a short note on whether prompt variants were tried would help readers reproduce the SFT baseline.","section":null},{"comment":"Figure 2 and IoU protocol: alignment/normalization follows CAD-Coder; a one-sentence restatement of the rigid alignment procedure would make the metric self-contained.","section":null},{"comment":"Table 1: “100 test subset for final evaluation same as [3]” is clear, but the exact sampling seed or list should be released with the code for exact reproducibility.","section":null},{"comment":"§4.2: GPT-5.2 is called “second best” on Fusion; ensure the same model version and decoding settings are fixed in the released evaluation scripts.","section":null},{"comment":"Appendix Figures 7–8 usefully show failure modes (thin sheets, non-enclosed volumes); cross-reference them earlier in §4 when discussing qualitative results.","section":null}],"recommendation":"major_revision","confidential_remarks":"The work is a solid empirical systems paper for a CAD/CE venue. The synthetic-proxy and small-subset issues are real but fixable within revision; they do not appear to reverse the ranking on the stated benchmarks. Fit is appropriate for a computational engineering / CAD journal; novelty relative to CAD-Coder is mainly the orthographic conditioning plus the open RL recipe rather than a new architecture."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"This is the paper that finally treats multi-view orthographic drawings (with dashed hidden lines) as the input for VLM CadQuery generation, instead of perspective renders or text. The practical pieces are the pythonOCC generator that turns STEP into first-angle drawings at scale (>150k from DeepCAD/Fusion) and the dual training recipe: SFT on GenCAD-Code where labels exist, then Dr.GRPO-style sequence-level RL with pure IoU reward when they do not.\n\nWhat works: 100% valid executable code on both 100-example test subsets, mean IoU 0.7922 (DeepCAD SFT) beating CAD-Coder’s 0.7361 and 0.5601 (Fusion RL) beating GPT-5.2’s 0.5181. Methods are standard and clearly written; the RL initialization from the SFT checkpoint is sensible; code and generator are promised open. No circularity in the metric—IoU is on held-out STEP solids after execution. Citation pattern is appropriate; they correctly note prior image-to-CadQuery and RL-for-CAD work while flagging that orthographics were under-served.\n\nSoft spots are scope, not internal cracks. Drawings only carry three bounding-box dimensions, sources are sketch-and-extrude only, evaluation is small subsets without error bars or real industrial scans, and transfer to messy full-dimensioned drawings remains untested (they own this in §3.1 and §5). Those limits do not reverse the ranking on the stated synthetic benchmarks.\n\nThis is for people building reverse-engineering or CAD-ML tools who need a reproducible orthographic baseline and an RL path when CadQuery labels are missing. It is not a theory paper. I would send it to peer review; the contribution is concrete enough and the evidence matches the claims. Engage if you care about drawing-to-editable-CAD; skim the generator and tables if you only need the numbers.","headline":"Clean empirical win on a real CAD pain point: orthographic rasters to executable CadQuery via SFT+geometry RL, plus an open drawing generator that actually includes hidden lines and dimensions.","tokens_in":14851,"tokens_out":507,"would_cite":true,"duration_ms":13160,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"grok-4.5","headline":"A vision-language model can read raster orthographic drawings and write CadQuery code that rebuilds editable 3D CAD solids more accurately than prior systems.","keywords":["orthographic drawings","CadQuery","vision-language models","3D CAD reconstruction","supervised fine-tuning","reinforcement learning","parametric CAD","STEP models"],"falsifier":"Evaluate the released model on a held-out collection of real industrial multi-view raster drawings that contain missing views, full feature dimensioning, and operations beyond sketch-and-extrude, and check whether mean solid IoU still exceeds the best baseline while code validity remains near 100 percent.","tokens_in":14829,"feed_emoji":"📐","tokens_out":941,"duration_ms":19742,"temperature":0.7,"pith_summary":"Most design intent still travels as raster multi-view orthographic drawings, yet manufacturing and analysis need parametric, editable 3D CAD. This paper claims that a vision-language model can close that gap by emitting CadQuery programs that compile into solids. When ground-truth CadQuery exists, the model is fine-tuned by ordinary next-token supervision; when only STEP geometry exists, it is optimized by reinforcement learning whose sole reward is the intersection-over-union of the executed solid against the true solid. A new pythonOCC pipeline supplies the missing training images—first-angle projections with dashed hidden lines and bounding-box dimensions—at scale from public STEP repositories. On DeepCAD and Fusion 360 evaluation subsets the resulting model produces 100 percent valid executable code and raises mean solid IoU by more than 7 percent relative to the next-best baseline. If the approach holds, the image drawings already circulating in industry could become direct sources of editable CAD.","feed_headline":"VLM turns orthographic drawings into editable 3D CAD","feed_subtitle":"SFT plus geometry-rewarded RL yields 100% valid CadQuery and over 7% higher solid IoU.","key_machinery":"Dual-regime training of a VLM: supervised fine-tuning on paired orthographic drawings and CadQuery when labels exist, followed by Dr. GRPO-style sequence-level reinforcement learning driven only by solid IoU rewards when labels are absent, enabled by a pythonOCC generator of standardized first-angle drawings.","core_discovery":"Ortho2CAD establishes that conditioning a modern vision-language model on three-view dimensioned orthographic drawings, then training it with supervised token likelihood where CadQuery labels exist and with sequence-level group-relative geometric IoU rewards where they do not, yields fully executable parametric CAD whose solids match ground-truth geometry more closely than open-source, closed-source, and domain-specialized image-to-CadQuery baselines.","pith_inferences":["The same geometry-reward loop could absorb drawings that carry full feature dimensions and section views once richer synthetic labels are available.","A hybrid inference path that uses the fine-tuned model for a strong first pass and a larger general model for iterative correction may close remaining IoU gaps on thin or multi-feature parts.","Because training data are restricted to sketch-and-extrude, new CadQuery corpora will be required before the method covers fillets, lofts, or freeform B-reps common in production.","Releasing the drawing generator lowers the barrier for others to test drawing-to-CAD pipelines on larger repositories such as ABC."],"forward_implications":["Where CadQuery labels exist, orthographic drawings become an effective conditioning signal for parametric CAD synthesis.","Where only STEP geometry exists, executable-geometry rewards can adapt a model without any target code.","Large orthographic drawing corpora can be generated automatically from any public STEP repository.","Code validity can reach 100 percent while mean solid IoU still improves over specialized and general VLMs.","The output remains editable CadQuery rather than meshes or voxels, so it stays usable in ordinary CAD workflows."],"fun_headline_variants":["Ortho2CAD VLM maps orthographic drawings to editable CadQuery code","SFT and geometry RL yield 100% valid CAD from three-view drawings","VLM reconstructs parametric solids from dimensioned orthographics","Ortho drawings to 3D CAD via VLM with 7% higher solid IoU","Vision-language model outputs executable CadQuery from raster views"],"cache_read_input_tokens":128,"weakest_assumption_plain":"Clean synthetic first-angle drawings that show only three bounding-box dimensions and come from sketch-and-extrude parts are assumed to be a good enough proxy for the incomplete, fully dimensioned raster drawings used in real manufacturing.","fun_headline_variants_meta":{"raw":{"variants":["Ortho2CAD VLM maps orthographic drawings to editable CadQuery code","SFT and geometry RL yield 100% valid CAD from three-view drawings","VLM reconstructs parametric solids from dimensioned orthographics","Ortho drawings to 3D CAD via VLM with 7% higher solid IoU","Vision-language model outputs executable CadQuery from raster views"]},"model":"grok-4.5","effort":"low","cost_usd":0.004982,"raw_usage":{"total_tokens":1457,"prompt_tokens":844,"num_sources_used":0,"completion_tokens":79,"cost_in_usd_ticks":49820000,"prompt_tokens_details":{"text_tokens":844,"audio_tokens":0,"image_tokens":0,"cached_tokens":256},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":534,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":844,"tokens_out":79,"duration_ms":4671,"temperature":1.0,"reasoning_tokens":534,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-13T05:59:26.415155+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Evaluate the released model on a held-out collection of real industrial multi-view raster drawings that contain missing views, full feature dimensioning, and operations beyond sketch-and-extrude, and check whether mean solid IoU still exceeds the best baseline while code validity remains near 100 percent.","supporting_citations":[],"review_version":1}