What are common challenges in creating mesh models? 8 measures to improve quality

When creating mesh models, appearance alone is not enough. In practice, quality is evaluated by factors such as geometric fidelity, dimensional reliability, minimal defects, and ease of handling in downstream processes. Therefore, even if a model appears fine right after creation, problems often surface when moving on to subsequent stages such as design verification, quantity assessment, as-built verification, maintenance, and documentation.

Especially in practical work where you must run the workflow from on-site measurement to point cloud processing, surface generation, finishing, and delivery within a limited time, the quality of mesh models is difficult to stabilize based solely on personal judgment. If each operator uses different criteria, the same object can result in variations in model density and smoothness, hole-filling policy, and handling of unwanted items, leading to inconsistencies in deliverables. As a result, rework increases, impacting schedules and reliability.

In this article, we first organize the typical issues that commonly occur in mesh model creation, and then explain eight measures to improve quality. Presented from a practical perspective and written clearly, this will be useful both for those who are preparing to establish operations for mesh model creation and for those who already have workflows in place but feel that quality is unstable.

Table of Contents

• Reasons why issues commonly arise during mesh model creation

• Improvement measure 1: Determine the necessary density and range during the measurement stage

• Improvement measure 2 Ensure that noise and unwanted objects are reliably reduced in preprocessing

• Improvement measure 3: Optimize the surface generation conditions for each object

• Improvement measure 4 Do not casually auto-complete missing data or fill gaps

• Improvement 5: Treat sharp edges and smooth surfaces separately

• Improvement 6: Solidify management of coordinate systems and scale early in the project

• Improvement measure 7 Balance lightweighting and reproducibility according to the application

• Improvement measure 8 Standardize inspection criteria and correction rules

• Summary

Reasons Why Problems Tend to Occur in Mesh Model Creation

The biggest reason the quality of mesh model creation tends to be unstable is that the appearance of the deliverable and its practical usefulness do not necessarily match. A model that looks smooth on screen may in fact have crushed details, messed-up face orientations, or be in a state that cannot be used for dimensional verification. Conversely, a model that looks somewhat rough may still retain the necessary shapes correctly and be perfectly usable in practice. In other words, the quality of a mesh model cannot be judged by appearance alone; consistency with its intended use is what matters.

Furthermore, a mesh model is not completed on its own; its quality is determined by the accumulation of multiple processes such as measurement conditions, alignment, point cloud quality, noise removal, surface-generation conditions, and finishing strategy. If any one process is pushed beyond what is feasible, subsequent processes will try to compensate, leading to an increase in unnatural corrections and, as a result, a decline in overall quality. For example, if surface generation is heavily adjusted when captured points are insufficient, the appearance may be filled in, but shapes that do not exist on the real object are likely to be created.

A common occurrence in practical work settings is that, in an effort to shorten working time, initial-stage checks are skipped. If you proceed while it is unclear which parts of the object should be prioritized, which parts can be somewhat rough, and to what extent defects can be tolerated, the person responsible will have to redo decisions repeatedly later. This not only increases the workload but also directly leads to variability in quality standards.

That is why improving mesh model creation is not simply a matter of adding more processing steps, but of clarifying what decisions are made at which stage. From here, we will organize common challenges into eight improvement measures and, in order, examine practical approaches that lead to quality improvement.

Improvement 1: Determine the required density and range during the measurement stage

When improving the quality of a mesh model, the first thing to review is the measurement stage. No matter how carefully you adjust things in post-processing, if the necessary information wasn't captured in the first place, you won't achieve a highly reproducible model. In practice, attention tends to focus on the processing conditions for surface generation, but the causes of poor quality often lie in the preceding stage.

A case that particularly tends to cause problems is insufficient acquisition density for the object. Even if it seems sufficient when looking only at flat surfaces, features such as steps, edges, openings, thin members, and transitions of curved surfaces become ambiguous if point sampling is coarse. When meshing is performed in that state, corners become rounded, boundaries blur, and surfaces that differ from the actual object are easily generated. If this issue is discovered after the work, re-measurement becomes necessary, resulting in the largest rework.

The key improvement is to work backward from the required accuracy for the intended use. The required density will differ depending on whether the model is a rough overview for viewing, used for shape verification, or used for dimensional measurement and comparative validation. The important thing is not to capture everything at high density, but to prevent information shortages in critical areas. In practice, it is better to keep the whole at a standard density while focusing on capturing only areas with large shape changes or areas that will be referenced in later processes.

Also, setting the capture area is important. If you prioritize only the object's outline and don't include enough background or connection points, it will become difficult later to separate parts or verify alignment. Conversely, if you capture too much of the surrounding area, removing unwanted items will take time. Deciding the area from three perspectives—the object's main body, the surrounding parts that show connection relationships, and the margins needed for alignment and cropping—helps stabilize downstream processes.

What’s needed at this stage is not advanced operational skills but the ability to draw up a measurement plan based on an understanding of which information will be used in downstream processes. If you truly want to improve the quality of the mesh model, you should consider that most of the work has already begun before you even enter the surface-generation screen.

Improvement 2: Ensure Reliable Reduction of Noise and Unwanted Objects in Preprocessing

Many of the disturbances in mesh models stem from moving on to surface generation before noise and unwanted objects in the point cloud have been sufficiently removed. When isolated points, reflection-induced spikes, moving people or vehicles, swaying vegetation, or unstable points caused by the measurement angle are mixed in, surfaces can be formed where they shouldn’t be, or the surface can become rough.

On-site, if you overdo noise removal you can erase necessary shapes, so there are situations where you want to process conservatively. That judgment itself is not wrong, but the problem is leaving the criteria vague and having operators process based on their own instincts. If one person errs on the side of caution while another cuts boldly, the final model’s overall character can change greatly even for the same subject. This leads to variability in quality.

To improve things, rather than treating preprocessing as a single bulk cleaning operation, it is effective to organize it by purpose. First, remove isolated points or extreme outliers that can clearly be judged as measurement errors. Next, separate out temporary reflections or ghosting that are not part of the target object. Then check the density and continuity of points that should remain as the object’s surface, and make local additional adjustments if necessary. Simply following this order will reduce both over-removal and under-removal.

The important thing is not to judge unwanted objects solely by appearance. For example, something that looks like a protrusion may actually be part of the equipment, and conversely a spot that appears to overlap the main body may be incidental noise. Processing based only on shape without understanding the structure of the object can easily lead to losing correct information. Whether the preprocessing personnel understand the object's intended use and shape characteristics has a significant impact on quality.

When the quality of noise reduction improves, subsequent surface generation becomes remarkably stable. Because you no longer need to force corrections during surface generation, unnatural stretching and loss of fine detail are reduced, and the time needed for fixes is shortened. Preprocessing may seem like a modest step, but carefully refining it is the quickest way to achieve high quality.

Improvement 3: Optimize surface generation conditions for each object

One common mistake in mesh model creation is applying the same surface generation settings to every object. In practice, processing settings are sometimes templated to improve efficiency, but if the target objects have different geometric characteristics, the appropriate settings will naturally differ. Structures with many flat surfaces and objects with organic curved surfaces require different triangle layouts and different approaches to preserving fine details.

For example, if you strongly prioritize smoothness for an object where you want to retain many fine bumps and depressions, characteristic shapes will be smoothed out. Conversely, if you prioritize detail preservation too much for an object composed mainly of large flat surfaces, unnecessary fine noise will remain on the surface, giving it a rough appearance. In other words, surface generation always involves trade-offs, and you need to decide what to keep and what to refine according to your objectives.

For improvement, it is important to abandon the assumption that surface generation must be completed in a single pass. First, create a rough version under baseline conditions and check for surface irregularities, edge collapse, reproduction of thin features, and the occurrence of holes. Then, adjust parameters such as density, the handling of nearby points, smoothness, and connection conditions according to the object, and iteratively refine while recording which changes produced which results. The accumulation of these comparisons leads to a reproducible workflow.

Also, trying to process the entire subject under a single set of conditions often runs into problems. When broad flat areas and fine details coexist, it’s more reasonable to separate the approach by region. If you try to establish settings that both stabilize the overall contours and preserve fine details at the same time and end up compromising, you will likely get results that satisfy neither. Adopting the idea of combining partial processes as needed makes it easier to balance quality and efficiency.

Optimizing surface generation conditions is not about fine-tuning numbers. It means deciding, based on the object's characteristics and intended use, which shapes should be prioritized and preserved. Once these decision criteria are clear, adjustments to processing conditions become less uncertain and the quality of the resulting product becomes more consistent.

Improvement 4: Do not automatically impute missing values or fill in gaps easily

When creating a mesh model, missing parts or holes reduce the perceived completeness, so you naturally want to fill them quickly. However, casually relying on automatic hole-filling can lead to the most serious quality degradation. That's because the filled surfaces are likely to be generated by algorithmic estimation rather than representing the actual object's shape.

In particular, openings, shadowed areas, overlapping parts, the back side, and thin connecting sections are prone to incorrect interpretation when filled in without sufficient information. Even if they appear neatly closed on the surface, areas that should be open may be blocked, or complex shapes may be simplified. While this may not be a major problem for viewing purposes, it can lead to misunderstandings when used for verification, comparison, or explanatory materials.

The basic approach to improvement is to first classify the reasons for missing data. The response differs depending on whether it is simply an acquisition omission, a structurally occluded area, a side effect of noise removal, or a problem with surface-generation conditions. If you fill gaps without examining the cause, it may appear resolved on the surface but the model’s reliability will decline. Before filling, it is important to make a habit of returning to the original point cloud and checking the surrounding information.

Also, even when filling in gaps is necessary, it is essential to make clear how much supplementation is permissible. For example, a minor touch-up to improve appearance should be handled differently from supplementation that amounts to a shape estimation and therefore requires accountability. In practice, deliverables are sometimes handed over with this line left ambiguous, but that leaves later reviewers uncertain about how much they can trust the work.

To improve quality, you should prioritize preserving the accuracy of information over eliminating gaps. For parts that require completion, it is important to handle them with that premise in mind. The goal is not to eliminate missing data entirely, but to refine the model within a range that does not undermine consistency with the real-world object; that is the proper measure of improvement.

Improvement 5: Treat sharp edges and smooth surfaces separately

Edge representation is extremely important as a factor that affects the quality of mesh models. A common practical issue is attempting to smooth the entire model uniformly and thereby rounding off corners and boundaries that should remain sharp. As a result, even if the appearance looks neat, the contours of the structure become softened and the shape's features are lost.

On the other hand, being overly focused on preserving edges can cause unnatural kinks to appear on curved or continuous surfaces. In other words, sharpness and smoothness are elements that should coexist, and applying only one of them uniformly across the entire object will compromise quality. This problem is especially likely to occur in objects where linear components and curved components coexist.

The approach to improvement is clear: treat separately the parts that should remain as corners and the parts that should be smoothed into continuous surfaces. For example, edges, corners, joints, notches, and step risers carry meaning for shape recognition, so it is important not to make them ambiguous. Conversely, gentle curved surfaces and fine undulations on broad faces are often better smoothed depending on the application, producing a more visually clear and easier-to-handle model.

What matters here is not simply making something look sharp. It is about understanding, based on the actual shape characteristics, which boundaries are meaningful. For example, with constructed elements and equipment, corners that affect dimensions or mounting positions can be important. If those parts become ambiguous, the accuracy of explanations and verification drops. Conversely, for natural objects or subjects that undergo aging, overly accentuating corners can stray from the actual condition.

As a workflow practice, it is effective to include a step to consciously check the contours that should be preserved before smoothing the whole model. Edge preservation is not a fine cosmetic adjustment but a process to preserve the model’s intended meaning. Simply adopting this perspective can significantly change the credibility of the completed mesh model.

Improvement Measure 6: Solidify management of coordinate systems and scale early in the workflow

If you focus only on appearance, it’s easy to overlook the management of coordinate systems and scale. Even if a mesh model has a well-formed shape, it becomes difficult to use in practice if the handling of position and dimensions is ambiguous. In particular, the reliability of coordinates and scale is indispensable when overlaying other data, performing time-series comparisons, reconciling with the site, or integrating the model into drawings and reports.

A common issue is that units and standards become mixed during the course of work. In one process, work proceeds under a local convention, while in another process it tries to align with external standards, and subtle discrepancies accumulate. These eventually manifest as positional or dimensional differences, producing deliverables whose shapes are correct but that are difficult to use. This kind of problem is especially likely to occur in projects involving multiple people.

To make improvements, it is necessary at the early stages of the work to decide which standard will be used to manage positions, which units will be employed, and where consistency checks will be performed. What matters more than everyone understanding the detailed theory is unifying the handling rules for each stage of the process. If the standard is clear, it becomes easier to maintain quality even when responsibility is transferred to another person midway.

Also, it's safer not to limit scale verification to the initial check. Verifying consistency at the milestones where data is transformed—after point cloud processing, after surface/mesh generation, after simplification, etc.—reduces the risk of noticing large discrepancies later. In practice, people often trace back the cause only after something feels off in the final stage, but that greatly increases the verification cost.

The quality of a mesh model is not just about the cleanliness of its surface. It includes what is where and the degree of certainty with which those things are represented. Establishing the coordinate system and scale early on forms the foundation that supports quality behind the scenes.

Improvement Measure 7: Balance weight reduction and reproducibility according to application

When working with mesh models in a professional context, issues of data size and display speed are unavoidable. Left at high density they become difficult to handle, and sharing, viewing, or incorporating them into reports can become burdensome. Therefore mesh simplification is necessary, but if taken too far it can remove shapes that should be preserved and lead to a decline in quality.

A common mistake is to treat lightweighting as simply data reduction. Reducing the face count will make processing lighter, but if you uniformly coarsen even important areas, contours will become distorted, details will vanish, and the model will become unusable for comparison and verification. Conversely, if you try to preserve everything, operability decreases and the scope of practical application narrows. In short, what matters here as well is optimization tailored to the intended use.

The key point for improvement is to decide in advance who will use the model and in what situations. If the model is intended mainly for viewing or sharing, it is effective to make it lighter while preserving its visual impression. Conversely, if the model will be used for shape verification or comparison, certain areas need to retain density. Even within the same project, it is practical to manage separate models—a preservation model that remains close to the original and a lightweight model for operational use.

Also, you should always perform a check after lightweighting. If you deliver the file without consciously comparing which features changed before and after the reduction, people who use it later may unintentionally misread it. In particular, changes such as straight lines appearing wavy, hole edges deteriorating, or thin features disappearing cannot be detected from file size numbers alone. Quality checks must always be carried out from both visual and geometric perspectives.

Lightweighting is not a process of lowering quality, but a design approach to enhance usability. By shifting to the mindset of reducing only unnecessary weight while preserving the required reproducibility, you move closer to mesh models that are useful in practice.

Improvement Measure 8: Standardize inspection criteria and correction rules

In mesh model creation, what ultimately causes differences in final quality is not only individual technical differences but also the presence or absence of inspection standards. If it is unclear which conditions are acceptable and what should be corrected, each person in charge can only judge based on their own experience. As a result, something that passes in one project may be sent back in another.

In practice, even when the work itself is proceeding, it is not uncommon for inspections to be dependent on individual personnel. For example, if criteria such as how much surface roughness is acceptable, how much defect is problematic, or to what extent residual foreign matter warrants rework are not shared, both operators and inspectors are left uncertain. That uncertainty increases back-and-forth corrections and destabilizes both delivery schedules and quality.

To improve this, it is effective to change mesh model inspection from a subjective evaluation to item-by-item checks. For example, viewing the model from perspectives such as whether the object's contours are preserved, whether there are missing parts in critical areas, whether there are unnecessary faces, whether scale and positional alignment are consistent, and whether the density is appropriate (neither excessive nor insufficient) for the intended use helps achieve more consistent judgments. This is not about creating an elaborate quality-control checklist, but about having a minimum common language.

Furthermore, correction rules should also be clarified. Organizing which problems should be returned to preprocessing, which should be fixed by readjusting surface-generation conditions, and which should be handled in finishing will reduce wasted trial-and-error. Operating by restarting from the beginning every time a problem occurs will make both time and quality inconsistent.

Standardization may sound rigid, but its purpose is not to constrain work—it is to make quality reproducible. If mesh model creation is to be run as an ongoing operation, the single most effective improvement is to establish processes that bring the results to a consistent level of quality regardless of who is responsible, rather than relying on the intuition of skilled individuals.

Summary

The challenges that arise in creating mesh models cannot be solved by surface-generation operations alone. It is necessary to consider how to obtain the required information, how to organize noise, which shapes to prioritize preserving, how far to allow filling-in, how to manage position and scale, and ultimately what criteria will be used to deem the result acceptable. The eight improvement measures introduced here may appear independent, but in fact they are strongly interrelated. If measurements are appropriate, preprocessing will be stable; if preprocessing is in order, surface generation will become natural; and if surface generation is stable, decisions about corrections and mesh simplification will also become easier.

What matters for practitioners is not producing a perfect mesh model in a single pass, but creating a workflow that can consistently deliver quality sufficient for its intended use. To achieve that, it is essential not only to pursue individual working techniques but also to identify where in the overall process quality is most likely to deteriorate and to standardize decision criteria. Reviewing outputs not only for visual tidiness but also from the perspectives of whether they can be used downstream without hesitation, whether they are easy to explain, and whether they are easy to compare will clarify the direction for improvement.

Furthermore, the quality of a mesh model is not determined solely by the creation process. Especially in outdoor measurements and tasks involving positional information, the reliability of the underlying positional data greatly influences the subsequent model quality. If on-site acquisition accuracy is unstable, you will struggle with consistency checks and comparison work no matter how carefully you process the data. Therefore, if you want to stabilize quality from the outset of measurement, it is effective to reassess the methods used to obtain positional information.

For example, for tasks that want to perform on-site positioning more easily and with higher accuracy, using LRTK, an iPhone-mounted GNSS high-precision positioning device, is an option. Because it makes it easier to proceed with photography and recording while recording the position on site, it can reduce the effort of checking positional relationships later and help streamline the data organization that forms the basis for mesh model creation. If you want to improve mesh model quality not only through post-processing but by getting things right from the on-site data acquisition stage, it is important to review the entire workflow, including improvements to accuracy at this input stage.