Explore

TEST_SCANS

Test 1 – TOOL COMPARISON

Comparison between RealityScan, Meshroom, and Metashape.

No texture comparison

Focus on processing time and output quality across the three tools

Capture ID: 08_09_2025_SCAN

Number of photos: 95

f-stop: f/1.4 f-stop should be around f/18 or max

focal lens: 50mm (good, no distortion)

⁠

⁠

TEST 1 _ RESULT NOTE

TEST 1 _ RESULT NOTE

Tools

Pros and Cons

REALITYSCAN

Automatic optimization process. Fills holes by generating rough geometry where data is missing

MESHROOM

Higher chance of crashes. Longest overall processing time

Metashape

Produces relatively better results. Leaves surfaces open where data is missing (no automatic hole filling)

There are no rows in this table

⁠

Photogrammetry Tools Comparison

Photogrammetry Tools Comparison

Feature

Metashape (Agisoft)

Meshroom (AliceVision)

RealityScan (Epic Games)

Platform

Win / macOS / Linux

Win / Linux

iOS (official), Android (beta)

Cost

Paid ($179–$3,499)

Free, open source

Free

Ease of Use

Medium – pro but technical

Harder – needs node tuning

Easy – mobile app

Performance

Fast, GPU-optimized, handles huge sets

Slow, crash-prone on big sets

Fast but limited by phone/cloud

Quality

High detail, industry standard

Good if tuned, less robust

Decent, simplified for games

Scale

Thousands of photos

~50–300 images typical

Up to 300 images

Best For

Surveying, VFX, GIS, pro assets

Hobbyists, research, experiments

Quick assets, AR/VR, Unreal/Sketchfab

There are no rows in this table

⁠

TEST 2 _ TOOLS COMPARISON 2

Comparison between RealityScan and Metashape.

Comparing processing time and output quality (geometry + textures)

Capture 2: 09_09_2025_SCAN

Scan Amount: 427 Photos

Image ID (below)

f-stop: f/4 f-stop should be around f/18 or max

TEST 2 _ RESULT NOTE

TEST 2 _ RESULT NOTE

Tools

Pros and Cons

REALITYSCAN

Algorithm automatically closes geometry holes to optimize the scan mesh, but results are not always ideal. Meshes tend to have more noise and appear rougher, both in geometry and in texture colors/details

METASHAPE

Produces relatively better results when sufficient input data is provided. No open holes in geometry. Delivers higher quality in both textures and performance

There are no rows in this table

⁠

⁠

⁠

⁠

EXTRA - RealityScan (HighQuality Mesh) v.s. Metashape

⁠

RealityScan (High-Quality Mesh)

RealityScan with the High-Quality Mesh option produces a much denser and more detailed mesh, offering higher resolution in geometry. Here are the result fomr NEMO-Test:

RealityScan HQ: 17,396,657 vertices

Metashape (result from two combined chunks of photo sets): 998,397 vertices

RealityScan - Pros

Potential for improved detail: RealityScan HQ ‘could’ probably achieve a much better results only if the input photos are of higher quality in both resolution and color, with sufficient coverage from all angles as well.

Automatic manifold handling: RealityScan HQ sometimes produces cleaner geometry. For example, in the Nemo test the base was automatically cut flat and potential holes were closed. However, this process can also introduce unintended artifacts, such as the abrupt cut at the top of the hair where the software was unable to interpret the surface correctly.

Free: no subscription or extra credits needed for HQ mesh generation

RealityScan - Cons

Increased noise: Higher density often introduces noticeable noise, clearly visible in the Nemo HQ and Snake HQ tests.

Heavy geometry: Dense meshes are harder to process, slow down editing, and complicate optimization workflows.

Texture quality: RealityScan generally produces weaker color textures compared to Metashape.

Artifacts in complex areas: Automatic cleanup can sometimes lead to unnatural cuts (e.g., trimming off hair details where geometry cannot be reconstructed).

RealityScan Stability (Hardware Consideration) RealityScan appears more sensitive to hardware limitations than Metashape. While it can leverage GPU acceleration, many processes remain CPU-intensive, and insufficient GPU memory often forces a fallback to CPU. As a result, RealityScan may show less stability in demanding workflows compared to Metashape’s more balanced CPU/GPU utilization. Therefore in our first two Nemo High-Quality Mesh tests, the weaker machine produced no output at all.

⁠

TEST 3 _ SCAN AMOUNT

In the previous scan, some data was missing on the top and bottom of the model. This caused the software to automatically generate grayish noise in the textures.

For Test 3, additional photos were taken from higher angles to cover the missing areas, along with more offset angles to provide extra input data and fill the gaps.

Capture: 298 images. (extra photos added to the previous scan chunk and re-generated for the Test 3 result)

Tool: Metashape

Single Chunk:

first chunk of photos captured in the TEST 2

Merged chunk:

new photos from the test 3 added and regenerated

⁠

⁠

TEST 3 _ RESULT NOTE

TEST 3 _ RESULT NOTE

Photo input

Pros and Cons

Single chunk

More missing data

Visible noise around the ears, lower chin, and in the hair

Merged Chunks

Better, higher quality results with more data input

Could not use ultra-high quality for textures. Longer processing time

There are no rows in this table

⁠

TEST 4 _ SNAKE SCAN

Part A - Metashape v.s. Reality Scan (Normal Detail Level)

The high-texture surface of the snake made it easy to scan, with no issues in surface capture and relatively low processing time.

However, due to the snake’s positioning, surfaces close to the plate were difficult to capture. This resulted in more noise and missing data, especially around the tail.

Capture: 247 images.

⁠

Mesh + Texture Comparation

Mesh + Texture Comparation

⁠

Mesh Comparation

Mesh Comparation

⁠

Metashape result close-up / no-tex

Metashape result close-up / no-tex

⁠

Metashape result close-up / tex

Metashape result close-up / tex

⁠

Part B - Metashape v.s. Reality scan (High Detail Level)

⁠

Test 4 RESULT NOTE

Test 4 RESULT NOTE

Photo input

Pros and Cons

Reality Scan

HQ mesh: more noisy in geometry and textures

Metasshape

better result, cleaner geometry and smarter in processing noisy data, result in decent quality. Better performance in colors as well.

There are no rows in this table

⁠

TEST 5 _ Realistic Human Texture recreation

Render Engine: Cycles 1080p

Textures created by: Human Pro add on + 3D Daz Genesis 9 character base mesh + UV map, Lighting in Blender

⁠

TEST_6_Sculp SCAN

Input: 102 images

F-stop: f/16

Focal Length 58mm

⁠

⁠

⁠

⁠

⁠

⁠

⁠

RESULT:

details are roughly captured, but result could be only used for props for the background.

To Improve:

Increase input photos – Capture more images (around 100–200 photos) with 70–80% overlap, and increase the distance between shots to better cover the upper areas, for example.

Add a cross-polarization filter to the Godox ring flash to achieve more even lighting and reduce reflections. This should improve the overall scan quality and provide greater flexibility for relighting later on.

TEST 6 _ Hand Scan using Reality Scan

Failed:

Due to too less inputs, bad alignment of cameras

Summary: skin-like silicone surfaces are very challenging to scan because of glossiness, translucency, and low texture detail.

one more reason cause the dirty out come is not only the hands texture but also the background shadows and colors. the Background color is breaking the scan by adding too much noises around

To Improve:

The object currently lacks surface texture, which makes it difficult for the software to align and reconstruct accurately. Adding more visible details or surface texture will improve scan quality. Alternatively, you can digitally sculpt the missing details afterward (see example below).

Avoid cropping the object — all input images should capture the entire object within the frame.

improve current background situation. color and shadows etc.

Background color suggestion