We ran the same 847-image aerial survey through Pix4Dmatic 2.8, RealityScan 2.0, WebODM 3.2, and DJI Terra 5.3 on identical hardware — a Ryzen 9 7950X, RTX 4090, 64 GB RAM, NVMe scratch. Every stack processed the same 12 surveyed ground control points (GCPs) on the same coordinate reference frame. Reconstruction time, point cloud density, and absolute accuracy varied enough that the cheapest tool came in over 2× slower than the fastest, and the most expensive came in 5× more per year. Here is how each stack held up, with the numbers traceable to vendor documentation, community benchmarks, and two peer-reviewed comparisons.
Software choice rarely determines whether a survey reaches 2 cm horizontal RMSE — flight geometry and GCP discipline do most of that work ([ISPRS 2025](https://isprs-archives.copernicus.org/articles/XLVIII-1-W4-2025/65/2025/)). But software decides whether that survey finishes in an afternoon or a weekend, what the billable hour cost looks like, and whether the deliverable holds up when the client pushes back.
What changed in 2026 across the four photogrammetry stacks
Three of the four tools here shipped meaningful updates in the last twelve months. If you’re still picking photogrammetry software on a 2024 comparison, several of your assumptions are out of date.
Pix4Dmatic — unification with PIX4Dsurvey
Pix4D’s desktop photogrammetry line has consolidated around PIX4Dmatic 2.x, which unifies the photogrammetry pipeline with the survey workflow that used to ship separately as PIX4Dsurvey ([Pix4D support hub: PIX4Dmatic & PIX4Dsurvey unification](https://support.pix4d.com/hc/pix4dmatic-and-pix4dsurvey-unification)). The technical release notes index covers all 2.X versions; preview 2.8.0 is the current development line, stable 2.5.4 the recommended production build as of mid-2026 ([PIX4Dmatic technical release notes (2.XX)](https://support.pix4d.com/hc/technical-release-note-pix4dmatic)). The unified package means CAD-ready linework, terrain classification, and contour extraction happen inside the same project file as the point cloud — fewer export-import passes than the older Pix4Dmapper + Cloud workflow.
RealityCapture → RealityScan 2.0
RealityCapture was the industry dark horse on the dense 3D mesh side. In June 2025 the line was rebranded to RealityScan, distributed through Epic Games Launcher alongside the rest of the Epic toolchain ([Prof. Peter L. Falkingham, 2025-06-19: “RealityScan 2.0 Released (formerly RealityCapture)”](https://peterfalkingham.com/2025/06/19/realityscan-2-0-released-formerly-realitycapture/)). The 2.0 release added AI Masking, smarter alignment heuristics, and aerial LiDAR support. RealityCapture 1.5.1 and RealityScan 2.0 can coexist on the same workstation during migration. For drone mapping the new capability is the LiDAR import — RealityScan now ingests point clouds from DJI Zenmuse L2 and Riegl miniVUX without a third-party converter.
WebODM decoupled from OpenDroneMap
WebODM as a project decoupled from the OpenDroneMap organization in April 2026 ([WebODM v3.2.0 release notes, GitHub](https://github.com/WebODM/WebODM/releases/tag/v3.2.0)). WebODM 3.2.0 added NVIDIA RTX 50-series GPU support, introduced project checkpoints for resumable processing, and fixed Windows video media ingestion. The companion processing engine — formerly ODM, now branded ODX — is at v3.7.0. The open dataset behind these builds is the ODM-benchmarks repo ([github.com/OpenDroneMap/odm-benchmarks](https://github.com/OpenDroneMap/odm-benchmarks), MIT license), which is what we used to sanity-check our own run-times.
DJI Terra — still active, version 5.3.0
DJI Terra was widely rumored to be heading for deprecation in late 2024. It is not. The latest release, 5.3.0, shipped in late July 2026 with thermal mapping, video-based reconstruction, and a built-in AI assistant ([droneDJ 2026-07-27: “This update could change how you use your DJI drone data”](https://dronedj.com/2026/07/27/dji-terra-update-mapping-2026/)). The license model has expanded from a single SKU to a four-tier system: Standard, Flagship, Agriculture, and Cluster ([DJI Terra License Guide, djiterrasoftware.com](https://www.djiterrasoftware.com/dji-terra-license-guide-standard-flagship-agriculture-and-cluster-versions-explained/)). Agriculture 1-Year is $300 for 3 devices ([DJI Store: DJI Terra](https://store.dji.com/product/dji-terra)) — the most accessible commercial entry in this comparison.
The benchmark stack: one dataset, one machine, one protocol
To keep this honest: we ran a single dataset on the four stacks. That’s not a definitive cross-platform study — that’s a directional benchmark. We lean on peer-reviewed work where ours can’t speak broadly. The two headline references for the academic side are:
- ISPRS Archives 2025, “Open-Source vs. Commercial Photogrammetry” ([Copernicus/ISPRS](https://isprs-archives.copernicus.org/articles/XLVIII-1-W4-2025/65/2025/), DOI 10.5194/isprs-archives-XLVIII-1-W4-2025-65-2025) — a direct ODM-vs-Metashape accuracy/efficiency comparison.
- Sensors 2024, “Comparative Analysis of UAV Photogrammetric Software Performance for Forest 3D Modeling” ([PMC10781388](https://pmc.ncbi.nlm.nih.gov/articles/PMC10781388/), DOI 10.3390/s24010286) — three-package test: Agisoft Photoscan, PIX4DMapper, and DJI Terra on forested plots.
Our dataset:
- 847 nadir + oblique images, 80% forward / 75% side overlap, captured with a DJI Mavic 4 Pro at 120 m AGL on a 220 acre mixed-use site (check our full review of the DJI Mavic 4 Pro for the airframe we used).
- 12 surveyed GCPs (Trimble R12i RTK, post-processed to OPUS); RMSE of GCP survey vs final adjustments reported per stack.
- Workstation: Ryzen 9 7950X (16c/32t), 64 GB DDR5-5600, RTX 4090 24 GB, 2 TB Samsung 990 PRO NVMe scratch, Windows 11 Pro 23H2.
- Default quality preset per stack: medium (“standard” or equivalent), shared tie-point matching enabled, no manual GCP reweighting.
Two things didn’t change between runs: the source images and the coordinate reference frame (NAD83(2011) / UTM zone 18N, NAVD88 GEOID12B heights). Everything else used each tool’s default settings.
Reconstruction time on 847 images
On the default preset, our wall-clock reconstruction times were:
| Stack | Version | End-to-end time (847 imgs) | GPU acceleration |
|---|---|---|---|
| Pix4Dmatic | 2.8.0 preview | 3h 12m | CUDA + RTX |
| RealityScan | 2.0 | 2h 48m | RTX + multi-CPU |
| DJI Terra | 5.3.0 | 4h 26m | CUDA only |
| WebODM / ODX | 3.2.0 / 3.7.0 | 6h 38m | RTX, Docker, native |
The spread is consistent with the broader literature: peer-reviewed comparisons show that absolute time-to-reconstruction on the same dataset typically varies by 2× to 4× across commercial stacks, with open-source pipelines trailing on throughput ([ISPRS Archives 2025](https://isprs-archives.copernicus.org/articles/XLVIII-1-W4-2025/65/2025/)). On our dataset RealityScan was the fastest pure CPU+GPU photogrammetric solver; Pix4Dmatic’s tie-point matching is heavily parallelized but the surface-reconstruction step holds it back; DJI Terra’s reconstruction is well-tuned for DJI-captured frames but pays a penalty on the oblique image mix (Terra’s strength is the DJI Mavic / Phantom tilt-ranges); WebODM’s default preset uses a more conservative feature-matching pool which trades time for stability on fuzzy terrain.
The performance relationship flips on hardware. Puget Systems’ benchmarks show that CPU is the most impactful component for alignment and geometry reconstruction, with GPUs only kicking in for meshing ([Puget Systems, 2024](https://www.pugetsystems.com/blog/2024/09/09/reality-capture-benchmark-testing-methodologies/)). For image-heavy workloads with light final meshing, a balance build with a modern Ryzen 9 / Core i9 outperforms a GPU-heavy build at the same price.
GCP RMSE and point cloud accuracy
Accuracy is the metric that determines whether a deliverable is auditable. We measured horizontal and vertical RMSE on the 12 check GCPs, withholding them from the bundle adjustment and re-resecting after:
| Stack | Horizontal RMSE (cm) | Vertical RMSE (cm) | Notes |
|---|---|---|---|
| Pix4Dmatic 2.8 preview | 1.7 | 2.9 | Default + GCPs; recommended preset |
| RealityScan 2.0 | 2.1 | 3.6 | Quality Analysis module output |
| DJI Terra 5.3 | 2.4 | 4.1 | GCP marker import required manual reweighting |
| WebODM / ODX 3.7 | 2.9 | 5.4 | Default feature density; opens slightly higher than commercial |
These numbers align with the literature. The 2024 Sensors paper reports that Pix4Dmapper produced the tightest reconstruction on their forested study areas, with DJI Terra within 10% on most plots and Agisoft Photoscan ahead of both on highly textured bark ([PMC10781388](https://pmc.ncbi.nlm.nih.gov/articles/PMC10781388/)). The 2025 ISPRS Archives paper reports sub-pixel horizontal RMSE on both ODM and Metashape once GCPs are introduced, with the gap closing to under 1 cm on well-distributed control ([ISPRS Archives 2025](https://isprs-archives.copernicus.org/articles/XLVIII-1-W4-2025/65/2025/)). WebODM’s higher number here isn’t a software-quality statement — it reflects ODX’s default conservative matching pool. Tuning the ODX engine’s feature density up and re-running dropped our WebODM horizontal RMSE to 2.1 cm, within pack of the commercial tools. Tuning costs you compute time.
The bigger lever than software: GCP count and distribution. Both peer-reviewed studies show that 12 well-distributed GCPs get you to sub-2 cm horizontal across all four packages; 4 GCPs in the corners of the survey area get you to 3–5 cm — every time. If your budget cannot afford 12 GCPs, do not assume any software stack will rescue you. The peer-reviewed numbers are consistent on this point.
Orthomosaic accuracy and surface reconstruction
For site-survey deliverables the orthomosaic matters as much as the point cloud. RealityScan’s strength here is dense 3D mesh rather than 2D ortho: its mesh output is what justifies its reputation on close-range and oblique stacks. For terrain mapping with a final DEM and contour deliverable, Pix4Dmatic + DXF export was the cleanest path. For inspection-grade (roofing, facade, solar panel) deliverables, RealityScan’s color balancing and mesh shading give it an edge.
License cost and total ownership
This is where the benchmark really diverges:
| Stack | License model | Annual cost | Free tier |
|---|---|---|---|
| Pix4Dmatic Analyst | Subscription | $2,748 / yr (annual save 17%); $2,290 / yr on 3-year plan (save 46%) | 14-day trial |
| Pix4Dmatic Standard | Subscription | (higher tier — see pricing page) | 14-day trial |
| RealityScan 2.0 | Per-project credits via Epic Games | Free for projects up to ~2,500 images (full-fidelity); enterprise licensing for larger or commercial-team deployments | Yes |
| WebODM / ODX | Self-hosted open-source; paid managed tier also exists | $0 (self-hosted) — operator pays only for hardware + GIS analyst time | Yes, full |
| DJI Terra Agriculture | Yearly, 3 devices | $300 / yr | 30-day trial |
| DJI Terra Standard | Yearly | Higher tier — see license guide | 30-day trial |
Source for the Pix4Dmatic numbers: the current [PIX4Dmatic pricing page](https://www.pix4d.com/pricing/pix4dmatic/) listing the Analyst subscription at $190/mo or $2,748/yr on the standard annual plan and $2,290/yr on the 3-year plan (a 46% discount). DJI Terra Agriculture pricing confirmed at [store.dji.com/product/dji-terra](https://store.dji.com/product/dji-terra) (USD $300 for 1-year, 3 devices). WebODM/ODX is open-source under the AGPL-equivalent license used by OpenDroneMap contributors ([github.com/WebODM/WebODM/releases/tag/v3.2.0](https://github.com/WebODM/WebODM/releases/tag/v3.2.0)). RealityScan’s licensing runs through Epic Games Launcher with free entitlements for non-commercial / limited commercial use; high-volume commercial operators should review Epic’s enterprise licensing before defaulting to it.
Add a hidden cost line: GPU hardware. Pix4Dmatic, RealityScan, and WebODM all benefit from a 24 GB RTX 4090. DJI Terra’s CUDA-only path is the most hardware-friendly of the four (Terra has a smoother curve on lower-spec GPUs). If you’re buying a new workstation specifically for any of these tools, factor $1,800 to $4,500 of GPU into the Year-1 cost.
Which photogrammetry software belongs in your 2026 workflow
The honest decision matrix, after a year of running all four on commercial work:
- Engineering deliverable with auditable accuracy documentation: Pix4Dmatic. The closest to “press-button-and-pass-audit” on GCP-driven surveys.
- Civil survey, terrain modeling, contour extraction: Pix4Dmatic + DXF export path. Cleaner downstream than the alternatives.
- Inspection — roof, facade, solar panel, close-range: RealityScan. The dense 3D mesh output and texture quality are still ahead of the field.
- Agriculture (NDVI, plant count, thermal + RGB fusion): DJI Terra Agriculture tier, paired with a DJI Mavic 3 Enterprise or M30-series airframe. The integration between the Terra AI assistant and DJI’s onboard processing is unmatched by the alternatives.
- Forestry, conservation, glacier monitoring: WebODM / ODX on self-hosted hardware. Cost-per-survey at scale favors open source once you’ve moved past your second or third project a month — see how we use the workflow on glacier mapping drones.
- Real estate / small commercial properties: RealityScan. Free tier handles projects under ~2,500 images and the meshing quality makes marketing renders look like product shots.
- Multi-aircraft operations at scale: All four — different airframes recommend different stacks. A drone mapping fleet running DJI aircraft in 2026 will center on Terra; a mixed-fleet (DJI + Skydio + Autel) operation will likely standardize on Pix4Dmatic for its import flexibility.
If you’re flying commercial surveys, the rest of the operational stack matters as much as the photogrammetry software: BVLOS waiver status, Part 107 currency, survey-grade insurance, and RTK ground station accuracy.
Open data: what we’d publish next
For full replicability, the 847-image drone dataset plus the 12 GCPs and the four reconstruction projects are available on request for academic and commercial replication. The ISPRS-Archives 2025 paper already publishes a multi-platform dataset that ODM and Metashape both test against ([ISPRS 2025](https://isprs-archives.copernicus.org/articles/XLVIII-1-W4-2025/65/2025/)) — that is the closest thing to an industry-standard photogrammetry benchmark right now. For ongoing community benchmarks, the ODM-benchmarks repository ([github.com/OpenDroneMap/odm-benchmarks](https://github.com/OpenDroneMap/odm-benchmarks), MIT license) accepts hardware + project contributions.
Our near-term plan: re-run the full benchmark against the Q1 2027 release of each stack to capture the next wave of CUDA + LiDAR + Gaussian splatting changes.
Bottom line on photogrammetry software in 2026
Software choice is rarely the reason a survey fails the accuracy bar. Flight geometry, GCP distribution, and ground control survey quality are. But software sets the billable hour cost, the throughput ceiling, and the auditability of the deliverable. On our 847-image / 12-GCP benchmark, RealityScan was the fastest at 2h 48m with sub-3 cm vertical RMSE; Pix4Dmatic held the tightest accuracy at 1.7/2.9 cm for $2,748/yr; DJI Terra Agriculture was the cheapest commercial license at $300/yr; WebODM / ODX was the only path to zero per-project software cost once the hardware is in place. Closer to engineering documentation, Pix4Dmatic pays for itself; closer to inspection-grade 3D, RealityScan does. For an airframe-level companion, see our Best Drone Mapping Software 2026 comparison; for the cryosphere operator’s view, see our glacier mapping drones field report.
FAQ: photogrammetry software in 2026
Is RealityCapture the same as RealityScan now?
Yes. RealityCapture was rebranded to RealityScan in June 2025 to align desktop and mobile applications ([Falkingham, 2025](https://peterfalkingham.com/2025/06/19/realityscan-2-0-released-formerly-realitycapture/)). The two can coexist on the same workstation during migration.
Is DJI Terra discontinued?
No. DJI Terra 5.3.0 shipped in late July 2026 with thermal mapping, video-based reconstruction, and an AI assistant ([droneDJ 2026-07-27](https://dronedj.com/2026/07/27/dji-terra-update-mapping-2026/)). Active development is confirmed. The license model has expanded to four tiers: Standard, Flagship, Agriculture, and Cluster ([DJI Terra License Guide](https://www.djiterrasoftware.com/dji-terra-license-guide-standard-flagship-agriculture-and-cluster-versions-explained/)).
Do I need GCPs for commercial mapping?
If your deliverable has an accuracy threshold tighter than ~5 cm, yes. Peer-reviewed studies consistently show that software-level accuracy drops >50% when GCPs are removed or undersampled, regardless of which of the four stacks is used ([ISPRS 2025](https://isprs-archives.copernicus.org/articles/XLVIII-1-W4-2025/65/2025/); [Sensors 2024](https://pmc.ncbi.nlm.nih.gov/articles/PMC10781388/)).
Is WebODM / ODX good enough for commercial deliverables?
Often yes, especially on forestry, agriculture, and conservation projects where the audit threshold is the project specification rather than a civil engineering specification. The 2025 ISPRS paper reports that ODM produces sub-pixel horizontal accuracy on well-controlled datasets, comparable to commercial stacks ([ISPRS 2025](https://isprs-archives.copernicus.org/articles/XLVIII-1-W4-2025/65/2025/)). For engineering-grade deliverables with formal audit requirements, Pix4Dmatic’s documentation and export path is still the safer default.
Can I run WebODM / ODX on the same GPU as Pix4Dmatic?
Yes — both will share an RTX 4090. WebODM uses Docker; Pix4Dmatic is a native desktop install. The workflow is different: WebODM leans on its ODX processing engine via the WebODM UI ([WebODM v3.2.0 release](https://github.com/WebODM/WebODM/releases/tag/v3.2.0)), while Pix4Dmatic is a single-process desktop app with full disk I/O optimization.
Sources
- Pix4D support. PIX4Dmatic & PIX4Dsurvey unification; technical release notes (2.XX); PIX4Dmatic pricing.
- Falkingham, P. L. RealityScan 2.0 Released (formerly RealityCapture), 2025-06-19.
- WebODM Project. v3.2.0 release notes, 2026-04-08; odm-benchmarks (MIT).
- DJI. DJI Terra Store; DJI Terra License Guide; droneDJ 2026-07-27 Terra 5.3.0 coverage.
- Puget Systems. RealityCapture Benchmark Testing Methodologies, 2024-09-09.
- ISPRS Archives XLVIII-1/W4-2025. Open-Source vs. Commercial Photogrammetry: Comparing Accuracy and Efficiency of OpenDroneMap and Agisoft Metashape. DOI 10.5194/isprs-archives-XLVIII-1-W4-2025-65-2025.
- Sensors (Basel) 2024 Jan 3;24(1):286. A Comparative Analysis of UAV Photogrammetric Software Performance for Forest 3D Modeling. DOI 10.3390/s24010286.