cs.GRAug 28, 2026

What Will This Copper Look Like Later? Forecasting Surface Appearance and Rendering It as a PBR Material

Authors: Teejuta Sriwaranon, Borworntat Dendumrongkul, Tanapat Chamted, Pizzanu Kanongchaiyos

Organizations: Department of Computer Engineering, Faculty of Engineering, Chulalongkorn University, Bangkok, Thailand · Faculty of Information Technology, King Mongkut’s Institute of Technology Ladkrabang, Bangkok, Thailand

Read PDF
Ask AI
Cite

Copy citation

DOINot available for this paper

Abstract

Digital design requires predicting how a metal surface will look later in its oxidation; this paper presents such a pipeline for copper. Given a fixed-camera observation, the system forecasts appearance 10 accelerated units ahead and converts it into the albedo, normal, roughness and metallic maps a renderer consumes. Forecasting is evaluated as an authoring tool would use it, on a copper specimen the system has not observed: an entire recording is held out, so training and checkpoint selection use one specimen and the test set is the whole of a second, recorded on a different day and condition. Under this protocol a learned spatio-temporal model with a monotone oxidation state, the most accurate forecaster within a single recording, is less accurate than copying the last observed frame on an unseen specimen, in both directions, as are three further trained architectures. The only forecaster that transfers is a closed-form global color extrapolation with no trained parameters, improving on copy-last-frame by 13.4% and 50.6%, with a margin that increases with horizon to +16.7% and +55.5% at t+10. Two controls qualify this: correcting every frame for the photometric drift measured on a non-oxidizing reference region leaves both margins intact, ruling out uncontrolled exposure as their source, and a moving-block bootstrap over the 6 independent windows each recording contains separates the larger margin from zero but leaves the smaller one not individually significant. The mechanism is measured: a learned susceptibility map encodes where corrosion begins on the training specimen and misleads on a new one, whereas the global color trajectory is what specimens share. The pipeline therefore deploys the closed-form forecaster for unseen specimens and the learned model only for continuing one already observed. Code, splits, protocol and leakage audit are released.

Figures & tables

Sourcet+1t{+}1t+5t{+}5t+Kt{+}Kmeanvs. persist. (%)
chamber →\rightarrow air
Persistence (copy last)3.42 e-51.71 e-44.58 e-42.14 e-40.00.0
affinergb3.22 e-51.54 e-43.81 e-41.85 e-4+13.4+13.4
patinanet4.21 e-52.31 e-46.93 e-43.06 e-4−43.2-43.2
predformer [ 16 ]1.21 e-49.54 e-44.92 e-44.28 e-4−100.0-100.0
convlstm [ 10 ]5.00 e-56.71 e-42.27 e-39.51 e-4−344.5-344.5
Table 1: Forecasting a copper specimen the model has never seen. Training and checkpoint selection use one recording; the test split is the entire other copper recording – a different specimen under a different condition. MSEfg\mathrm{MSE}_{\mathrm{fg}} ( ↓\downarrow ) at t+1t{+}1 , t+5t{+}5 , t+Kt{+}K and averaged over the horizon; the last column is the mean improvement over copying the last frame.
Figure 1: What the pipeline produces on a specimen it has never seen, followed far enough for the change to be visible. Over the evaluated horizon t+10t{+}10 this copper moves only 1.2%1.2\% of full scale, which the noise-calibrated metric resolves but the eye does not. By t+85t{+}85 it has moved 7.8%7.8\% and the darkening is obvious. Columns marked ∗ are extrapolation well beyond the horizon any model here was trained or evaluated for, and are shown to make the phenomenon legible and to expose the long-range behavior discussed in Section 8 , not as evidence of accuracy there. Where to look at t+10t{+}10 : the change is concentrated on the upper half of the plate, around the base of the mounting rod, where a gold-brown cast appears before the rest of the face darkens. The violet fringe along the upper-right edge and the overall shift toward deep red belong to the starred columns and are not present at the evaluated horizon.
Figure 2: The reverse transfer direction: trained on the open-air specimen, forecasting the chamber one. Same layout as Figure 1 , and the same caveat on the starred columns. Reporting both directions matters because the two recordings differ in difficulty (the open-air one is roughly three times noisier), and a single direction could be read as a property of one specimen rather than of the method. Where to look at t+10t{+}10 : the earliest visible change is on the lit left band of the plate, which warms toward orange, while the shadowed right half darkens slightly. The blue-black cast that dominates the right half in the starred columns is well beyond the evaluated horizon.
Figure 3: Forecast error against horizon on the held-out copper specimen, log scale, with the copy-last-frame reference pinned. Left and right are the two transfer directions.
Figure 4: The forecaster is not tracking the camera. Left: mean RGB of the specimen (solid) and of a non-oxidizing reference region — neutral backdrop, eroded away from the specimen and its mount — over each whole recording. In the chamber the reference is flat to 0.0010.001 while the specimen moves 0.1280.128 ( 105×105\times ); in open air the reference brightens by 0.0260.026 against a specimen change of 0.1190.119 . Right: affinergb ’s improvement over persistence before and after every frame is corrected for the drift measured on that reference region, with 95%95\% moving-block bootstrap intervals over the 66 independent 140140 -frame blocks each recording contains. The advantage survives correction in both directions.
Design decisionchamber →\rightarrow airair →\rightarrow chamber
context frames observed before forecasting
n=3n=32.19 e-47.38 e-5
n=5n=5 †1.72 e-45.43 e-5
n=7n=71.66 e-44.82 e-5
n=10n=101.43 e-44.68 e-5
persistence reference ( n=5n{=}5 windows)2.06 e-41.07 e-4
Table 2: Characterizing the shipped forecaster on the held-out specimen, MSEfg\mathrm{MSE}_{\mathrm{fg}} ( ↓\downarrow ), both transfer directions. One axis varies per block; †\dagger marks the configuration used everywhere else in this paper. Nothing here is trained, so the whole sweep is a few minutes of forward passes. The sweep samples 60 windows per configuration and the usable window set shifts with the context length nn , so the persistence references here differ slightly from Table 1 , which averages every test window at n=5n{=}5 .
Figure 5: Why the learned forecaster does not transfer. Left: the susceptibility map ρ(x)\rho(x) the model uses to decide where corrosion grows. Right: where that specimen actually changed, measured directly from its frames. On the specimen the model was trained on, ρ\rho tracks the real pattern ( r=0.68r=0.68 ). On the held-out specimen it is uncorrelated ( r=0.00r=0.00 ). Note the two specimens also differ in pose and framing, so this is evidence that a fitted spatial prior does not transfer, not that nucleation chemistry is unlearnable.
 +5+5+10+10+20+20+50+50
Texture sourceF LIPLPIPSF LIPLPIPSF LIPLPIPSF LIPLPIPS
affinergb0.2530.2080.2300.1630.2440.1880.3110.220
patinanet0.3280.2890.2770.1950.2610.1890.2440.185
Static control (canvas only)0.4450.3800.4530.3880.4320.3630.3410.289
Table 3: Rendering the forecast on the held-out specimen. F LIP and LPIPS ( ↓\downarrow ) between renders driven by predicted textures and renders driven by the true future frames, identical geometry and lighting. The static control is the conventional time-invariant material; both forecasters beat it at every horizon on both metrics.

Explore similar work

CardsList
Aug 31, 2026cs.LG

Generative artificial intelligence for reliable mechanistic reasoning for corrosion

Corrosion accounts for approximately 4% of global GDP, and reliable prediction is essential for timely mitigation. Machine learning effectively predicts corrosion rates from composition, microstructure, and environmental variables, but cannot explain the underlying mechanisms. A reliable approach in safety-critical materials engineering requires not only accurate retrieval but also mechanistically defensible reasoning, a capability that existing factuality metrics cannot assess. This work presents a domain-adapted retrieval-augmented generation framework for corrosion knowledge synthesis, demonstrated on magnesium alloy corrosion. Three open-weight language models (Llama-3.1-8B, Qwen-2.5-7B, Mistral-7B) are fine-tuned on 3,309 expert-verified question-answer pairs from 840 peer-reviewed papers and integrated with a hybrid dense-lexical retrieval pipeline. Retrieval augmentation produces Token F1 gains of 143-194%, with system faithfulness of 0.964 and context recall of 0.988. Blind external validation on newly published literature and in-house electrochemical data confirms trend-level generalisation. Reason Map, a proposition-graph framework, is further introduced; it independently constructs directed evidence graphs from generated answers and retrieved literature, enabling systematic detection of causal direction inversions and unsupported inferential leaps that flat factuality metrics cannot expose. The modular architecture can be applied across domains, offering a generalizable blueprint for trustworthy AI-assisted knowledge synthesis to circumvent corrosion, which can also be applied to other engineering domains.

M N Bharath, R K Singh Raman, Alankar Alankar

Department of Mechanical and Aerospace Engineering, Monash University, Clayton 3800, Victoria, Australia · Center for Machine Intelligence and Data Science, Indian Institute of Technology Bombay, Mumbai, Maharashtra, 400076, India

Retrieval-Augmented GenerationCausal Reasoning in Language ModelsMaterials ScienceEvidence-Grounded Generation
AskPDF
Jul 23, 2026cs.CV

Future Rendering ≠\neq Future Surface: A Benchmark and Dataset for Dynamic Surface Reconstruction Beyond the Observed Window

Dynamic-scene reconstruction is almost always evaluated inside the observed time window, yet deployment settings such as AR overlays, robot interaction, and anticipatory planning need the future surface: the geometry at times beyond those captured. No standard benchmark measures this. We introduce FutureSurf, a controlled diagnostic benchmark and dataset for future-time surface reconstruction that trades scene diversity for exact future ground truth and falsification controls. A method trains on the observed first 75% of a sequence; we score its extracted per-frame surface on the held-out future by Chamfer distance, reporting absolute future CD as the primary score and the future/observed gap as a diagnostic. The dataset contains eight analytically defined controlled motions, including three falsification controls, with exact per-frame ground-truth meshes. We also provide a ground-truth-side recoverability oracle. The release includes split files, scoring code, a benchmark card, and Croissant metadata. On the controlled motions, the DG-Mesh backbone leaves a 2.7-4.1×\times gap even for futures predictable in principle (four of five recoverable from observed motion by a fixed rule), while the falsification controls behave as designed (the surface-invariant motion shows no gap). Beyond the contributed dataset, the gap persists across six animated DG-Mesh asset scenes and a second backbone, Deformable-3DGS (2.0-6.6×\times; both share a deformation-MLP temporal model). The benchmark also shows that future rendering quality and future-surface accuracy are statistically decoupled, so the novel-view-synthesis metrics the field reports do not track future geometry. The future error is structured, concentrating where the surface moves. The dataset, evaluation toolkit, and scoring code are available on Hugging Face and GitHub (https://github.com/Ricky-S/futuresurf).

Yukun Shi, Minglun Gong

University of Guelph

Dynamic Scene Reconstruction3D Mesh Reconstruction
AskPDF
Apr 21, 2026cond-mat.mtrl-sci

Multimodal Transformer for Sample-Aware Prediction of Metal-Organic Framework Properties

Metal-organic frameworks (MOFs) are a major target of machine-learning-based property prediction, yet most models assume that a single framework representation maps to a single property value. This assumption becomes problematic for experimental MOFs, where samples reported as the same framework can exhibit different properties because of differences in crystallinity, phase purity, defects, and other sample-dependent factors. Here we introduce Experimental X-ray Diffraction Integrated Transformer (EXIT), a multimodal transformer for sample-aware prediction of MOF properties that combines MOFid with X-ray diffraction (XRD). In EXIT, MOFid encodes MOF identity, whereas XRD provides complementary information about the experimentally realized sample state. EXIT is pre-trained on one million hypothetical MOFs with simulated XRD to learn transferable representations, leading to improved downstream performance relative to existing approaches. EXIT is fine-tuned on literature-derived experimental datasets for surface area and pore volume prediction. Incorporating experimental XRD improves predictive performance relative to models without experimental XRD, and attention analysis and sample-level case studies further show that EXIT assigns different predictions to samples sharing the same MOF identity when their XRD patterns differ. These results establish a practical step from framework-aware to sample-aware MOF property prediction and highlight the value of incorporating experimental characterization into porous materials informatics.

Seunghee Han, Jaewoong Lee, Jihan Kim

Department of Chemical and Biomolecular Engineering, Korea Advanced Institute of Science and Technology, Daejeon 34141, Republic of Korea. · Department of Materials Science and Engineering, Korea Advanced Institute of Science and Technology, Daejeon 34141, Republic of Korea.

Multimodal PretrainingMaterials Property Prediction
AskPDF