DelAnyFlow: Unified Delineation Across Domains

• DelAnyFlow is a family of methods that employ a consistent delineation principle across varied systems, including diagrammatic models, optical-flow estimators, and remote sensing pipelines.
• The optical flow variants enhance segmentation by integrating frozen SAM features with FlowFormer or ARFlow architectures to reduce motion fragmentation and improve flow accuracy.
• In agricultural remote sensing, DelAnyFlow leverages YOLOv11-based instance segmentation and structured post-processing to extract topologically consistent field boundaries at large scales.

Delineate Anything Flow (DelAnyFlow) is a label that refers to multiple technically distinct constructs in the cited literature rather than to a single universally fixed method. In the 2025 agricultural remote-sensing literature, DelAnyFlow denotes a resolution-agnostic pipeline for large-scale field boundary mapping that couples the DelAny instance-segmentation model with structured post-processing, tile merging, and vectorization to produce topologically consistent vector boundaries (Lavreniuk et al., 17 Nov 2025). In other supplied technical summaries, the same label is also used for two-frame optical-flow systems that augment a flow backbone with Segment Anything Model (SAM) features or SAM-guided training mechanisms to reduce motion fragmentation (Zhou et al., 2023, Yuan et al., 2024). A further, broader usage appears in a diagrammatic adaptation of Unified Control and Data Flow Diagrams, where “DelAnyFlow” names a generalized notation for describing arbitrarily complex systems through unified control flow and data flow (Polkovnikov, 2016). Taken together, these sources establish DelAnyFlow as an overloaded term spanning diagrammatic system representation, dense motion estimation, and agricultural field delineation.

1. Terminological scope and historiography

The name DelAnyFlow is not introduced in a single canonical paper across all domains represented here. Instead, the supplied corpus documents at least three distinct usages.

In the software- and systems-modeling context, the term is attached to a generalized application of Unified Control and Data Flow Diagrams (UCD). The underlying method describes systems through control-flow relations, data-flow relations, strict system boundaries, and explicitly separated timelines, and the summary states that the same conventions can be used to build a “Delineate Anything Flow” diagram for arbitrarily complex systems (Polkovnikov, 2016). In this usage, DelAnyFlow is a diagramming language or representational scheme rather than a learned model.

In optical flow, one supplied summary explicitly recasts the architecture of SAMFlow as a recipe for building “Delineate Anything Flow (DelAnyFlow),” defined as a two-frame optical-flow estimator that plugs a frozen SAM image encoder into FlowFormer, fuses and adapts the resulting features via the Context Fusion Module (CFM) and Context Adaptation Module (CAM), and trains under standard optical-flow losses to eliminate fragmentation (Zhou et al., 2023). Another summary identifies “DelAnyFlow or UnSAMFlow” with an unsupervised optical-flow system based on ARFlow, augmented by a semantic augmentation module, a homography-based smoothness loss, and an optional mask-feature branch using SAM masks (Yuan et al., 2024). These two optical-flow formulations are not identical architectures.

In agricultural remote sensing, DelAnyFlow is presented as a turnkey, resolution-agnostic pipeline for extracting topologically consistent agricultural field boundaries at regional to national scales. Its core is DelAny, an instance-segmentation model based on the YOLOv11 family and trained on FBIS-22M, followed by a five-stage post-processing and vectorization pipeline (Lavreniuk et al., 17 Nov 2025). The related DelAny paper also uses “DelAnyFlow” to denote the end-to-end field-boundary delineation pipeline built from normalization, YOLOv11-style instance segmentation, and contour extraction (Lavreniuk et al., 3 Apr 2025).

A common misconception is that DelAnyFlow denotes a single architecture across domains. The supplied sources do not support that interpretation. A more accurate reading is that DelAnyFlow functions as a reusable label for “delineate anything” workflows in multiple technical settings.

2. Diagrammatic DelAnyFlow from unified control and data flow

In the UCD-derived usage, DelAnyFlow is grounded in an explicit formalization of process blocks, data holders, and two relations: $CF \subseteq P \times P$, the control-flow relation, and $DF \subseteq (P \cup D) \times (P \cup D)$, the data-flow relation (Polkovnikov, 2016). The summary defines

$$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$

and

$$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$

Control flow is denoted by a thin, straight, angled arrow with an optional timeline token, while data flow is denoted by a thick, curved arrow (Polkovnikov, 2016).

The graphical vocabulary is explicit. Rectangle denotes a process or code block; rectangle with double-struck vertical edges denotes a module, class, file, library, or equipment; ellipse denotes a data holder; wavy-bottom shape denotes a file or database; paper shape denotes a document or human-generated report; rhombus denotes an algorithmic decision; and a small rectangle labeled “OR” denotes a logical choice point (Polkovnikov, 2016). Additional relations include “has” or part-of, alias or type-of via the “is” line, and labeled create/delete operations (Polkovnikov, 2016).

A central construct is the timeline. The summary defines a timeline as the mechanism by which control is distributed and states that it is always drawn as a thin straight or angled arrow carrying a small rectangle labeled with a sequence number or timestamp. A process may emit a new timeline arrow to denote a sub-timeline, allowing arbitrary nesting depth (Polkovnikov, 2016). This explicit separation between blocks of code and timelines is presented as the means by which nesting of the control flow can be shown “as deep as necessary” (Polkovnikov, 2016).

The supplied examples illustrate the intended generality. One example represents a simple C++ function call using timeline tokens, data-flow arrows for variable passing, and return control flow; another represents a production-line station flow with module boundaries, a left-to-right control timeline, and data updates to a widget plus database writes (Polkovnikov, 2016). The adaptation rule is stated directly: use the identical palette of shapes and lines to model any system, with control flow interpreted as “resource/time allocation,” data flow as “material/information transfer,” modules as “subsystems,” and data holders as “buffers” or “containers” (Polkovnikov, 2016).

This suggests that the diagrammatic DelAnyFlow is best understood as a unifying notation for heterogeneous systems rather than as a domain-specific algorithm.

3. DelAnyFlow in supervised optical flow via SAM-enhanced FlowFormer

One optical-flow formulation of DelAnyFlow builds on FlowFormer by adding the frozen image encoder from SAM (Zhou et al., 2023). Each input frame $I \in \mathbb{R}^{H \times W \times 3}$ is processed by a SAM image encoder $E_S$, described as a ViT-backbone with frozen weights, producing high-level object-aware features $\Phi_S$, and by a FlowFormer context encoder $E_F$, described as a lightweight CNN producing low-level flow-context features $\Phi_F$ (Zhou et al., 2023). These are merged in the Context Fusion Module into $\Phi_C$, refined in the Context Adaptation Module into $DF \subseteq (P \cup D) \times (P \cup D)$0, and then passed to the standard FlowFormer correlation-decoder and update-block stack to produce the final optical flow field $DF \subseteq (P \cup D) \times (P \cup D)$1 (Zhou et al., 2023).

The Context Fusion Module concatenates SAM and FlowFormer features and passes them through two residual conv-blocks. The summary specifies that $DF \subseteq (P \cup D) \times (P \cup D)$2, with $DF \subseteq (P \cup D) \times (P \cup D)$3 denoting standard $DF \subseteq (P \cup D) \times (P \cup D)$4 convolutions and $DF \subseteq (P \cup D) \times (P \cup D)$5 a $DF \subseteq (P \cup D) \times (P \cup D)$6 depthwise convolution. The result is a fused feature tensor $DF \subseteq (P \cup D) \times (P \cup D)$7 carrying both integrity-aware object features from SAM and fine-grained local flow cues from the original encoder (Zhou et al., 2023).

The Context Adaptation Module addresses the mismatch between SAM pre-training for segmentation and the optical-flow objective. It introduces $DF \subseteq (P \cup D) \times (P \cup D)$8 Learned Task-Specific Embedding tokens $DF \subseteq (P \cup D) \times (P \cup D)$9, randomly initialized and learned end-to-end, and applies one or more Two-Way Attention blocks consisting of self-attention on $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$0, cross-attention from $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$1 to $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$2, an MLP-based embed-update, and cross-attention from $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$3 to $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$4 to yield $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$5 (Zhou et al., 2023). Positional encoding is cropped from SAM’s original positional encoding to match feature-map size (Zhou et al., 2023).

Training is described as largely following FlowFormer’s two-stage regimen, while the summary actually enumerates three stages and states that SAM is frozen at all times. Stage 1 uses FlyingThings3D only for 120 K steps with batch size 3, crop size $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$6, random scale in $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$7, random horizontal/vertical flips, AdamW with $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$8, $$CF = \{(p_i,p_j)\mid \text{control passes from } p_i \text{ to } p_j\},$$9, $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$0, one-cycle learning-rate decay, and gradient clipping at 1.0. Stage 2 uses a data mixture of FlyingThings3D, MPI-Sintel, KITTI-15, and HD1K for 240 K steps with batch size 3 and grad-accum $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$1, with $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$2. Stage 3 uses KITTI-15 only for 50 K steps with batch size 3 and the same learning rate. Hardware is listed as 3 RTX 3090 GPUs with mixed-precision training (Zhou et al., 2023).

The loss terms are the Endpoint Error

$$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$3

and an optional Robust Charbonnier / $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$4 term with $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$5, with total loss $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$6 or plus $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$7 if desired (Zhou et al., 2023).

Quantitatively, the summary reports on Sintel(train) and KITTI-15(train) that the FlowFormer baseline obtains $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$8 EPE on Sintel clean/final and $$DF = \{(x,y)\mid \text{information flows from } x \text{ to } y\}.$$9 EPE / $I \in \mathbb{R}^{H \times W \times 3}$0 F1-all on KITTI, whereas DelAnyFlow obtains $I \in \mathbb{R}^{H \times W \times 3}$1 EPE on Sintel and $I \in \mathbb{R}^{H \times W \times 3}$2 EPE / $I \in \mathbb{R}^{H \times W \times 3}$3 F1 on KITTI, corresponding to reductions of $I \in \mathbb{R}^{H \times W \times 3}$4 and $I \in \mathbb{R}^{H \times W \times 3}$5, respectively (Zhou et al., 2023). On Sintel(test) and KITTI-15(test), the summary gives FlowFormer++ at $I \in \mathbb{R}^{H \times W \times 3}$6 EPE clean/final and $I \in \mathbb{R}^{H \times W \times 3}$7 F1, and DelAnyFlow at $I \in \mathbb{R}^{H \times W \times 3}$8 and $I \in \mathbb{R}^{H \times W \times 3}$9 F1 (Zhou et al., 2023). It further states that the method yields an 18–19% EPE drop in occluded regions (Zhou et al., 2023).

The stated mechanism is that fragmentation in flow arises when models rely on purely local photometric cues, while SAM’s ViT encoder encodes global object integrity across occlusions, texture-poor regions, or brightness changes; CFM propagates full-object awareness to pixel context features, and CAM injects task-specific motion priors via LTSE tokens and cross-attention (Zhou et al., 2023).

4. DelAnyFlow in unsupervised optical flow via SAM-guided ARFlow

A second optical-flow usage equates DelAnyFlow with the UnSAMFlow formulation, an unsupervised network guided by SAM (Yuan et al., 2024). Here the backbone is an encoder-decoder adapted from ARFlow, not FlowFormer. The encoder constructs a 6-level fully convolutional pyramid $E_S$0, where level $E_S$1 has spatial size $E_S$2 and channel dimension 32, with all convolutions $E_S$3, stride 2 on downsample layers and stride 1 otherwise (Yuan et al., 2024).

The iterative decoder begins from the coarsest estimate $E_S$4 and, for $E_S$5, upsamples the previous estimate, warps $E_S$6, computes a $E_S$7 correlation window yielding an 81-dimensional cost, concatenates the upsampled flow, the cost volume, and a $E_S$8-convolved feature tensor, then predicts a residual flow through a small Res-block flow estimator and further refines it through a context net (Yuan et al., 2024). Final full-resolution flow is obtained by upsampling $E_S$9 by factor 4 with learned convex weights (Yuan et al., 2024).

An optional mask-feature module inputs SAM masks $\Phi_S$0. The masks are converted to a full segmentation $\Phi_S$1 by sorting masks by ascending area, assigning overlapping pixels to the smallest-mask identifier, and labeling uncovered pixels as background (Yuan et al., 2024). At each pyramid level, the method downsamples $\Phi_S$2, applies a $\Phi_S$3 convolution to image features, computes regionwise max-pooled feature representatives $\Phi_S$4, broadcasts them back to all pixels of region $\Phi_S$5, concatenates regionwise and local features, and applies another $\Phi_S$6 convolution to obtain mask features $\Phi_S$7, which are then warped and correlated like the image features (Yuan et al., 2024).

Two SAM-guided training mechanisms distinguish this formulation. The semantic augmentation module imposes self-supervised consistency under transformation using known augmentations $\Phi_S$8, including photometric transforms, 2D affine transforms, and semantic copy-paste of key-object masks from different samples. The augmentation loss is

$\Phi_S$9

with key objects selected to have large mask-overlap (Yuan et al., 2024). The homography-based smoothness loss replaces traditional second-order smoothness with a regional approximation: for each SAM segment region $E_F$0, a homography $E_F$1 is fitted by RANSAC if the inlier rate is at least 50%, and the regionwise loss is

$E_F$2

where $E_F$3 is the regional warp induced by $E_F$4 (Yuan et al., 2024).

The full objective is

$E_F$5

with $E_F$6 and $E_F$7, where $E_F$8 is an ARFlow-style photometric loss combining $E_F$9, SSIM, and Census terms (Yuan et al., 2024). Implementation details include PyTorch, the ViT-H “default” SAM model, KITTI raw and Sintel raw pretraining followed by fine-tuning, resizing to $\Phi_F$0 for KITTI and $\Phi_F$1 for Sintel, batch size 8, Adam with $\Phi_F$2, constant learning rate $\Phi_F$3 for 100 K iterations followed by OneCycleLR up to $\Phi_F$4 for 100 K iterations, and delayed activation of semantic augmentation and homography loss until 150 K iterations (Yuan et al., 2024).

Reported results on KITTI-2015(test) show Fl-all improving from $\Phi_F$5 for the ARFlow backbone baseline to $\Phi_F$6 for the full model with augmentation, homography, and mask features (Yuan et al., 2024). On Sintel Final(test), EPE improves from $\Phi_F$7 px to $\Phi_F$8 px (Yuan et al., 2024). Cross-domain results are also reported: training on KITTI and testing on Sintel Final yields EPE $\Phi_F$9 px with SAM versus $\Phi_C$0 px without SAM, and training on Sintel then testing on KITTI yields Fl-all $\Phi_C$1 with SAM versus $\Phi_C$2 without SAM (Yuan et al., 2024). Runtime is given as $\Phi_C$3 s on $\Phi_C$4 inputs on Tesla P100, excluding SAM mask extraction, with only $\Phi_C$5 M added parameters for the mask-feature module (Yuan et al., 2024).

The existence of both the FlowFormer-based and ARFlow-based optical-flow variants is another indication that “DelAnyFlow” is not a uniquely specified optical-flow architecture in the supplied materials.

5. DelAnyFlow for agricultural field boundary detection

In agricultural remote sensing, DelAnyFlow denotes an end-to-end methodology for country-level field boundary detection from satellite imagery (Lavreniuk et al., 17 Nov 2025). The central idea is to reframe field-boundary extraction as instance segmentation of individual field objects, then apply structured post-processing to transform raw instance masks into seamless vector polygons (Lavreniuk et al., 17 Nov 2025, Lavreniuk et al., 3 Apr 2025).

The DelAny model at the core of this pipeline is based on the YOLOv11 instance-segmentation family, specialized for multi-resolution satellite imagery (Lavreniuk et al., 17 Nov 2025). The full model uses a YOLOv11 backbone with a $\Phi_C$6 width multiplier, capped at 512 channels, yielding 379 layers and approximately 62 million parameters. DelAny-S is described as a lightweight variant with half the depth and one-quarter the width, 203 layers, and approximately 2.9 million parameters (Lavreniuk et al., 17 Nov 2025). The heads include a single-category class head for “field” using sigmoid activation and focal-style classification loss, a box-regression head using CIoU loss, and a mask head producing $\Phi_C$7 instance masks upsampled to tile resolution (Lavreniuk et al., 17 Nov 2025).

The related DelAny technical summary presents the pipeline as native-resolution image input, per-channel normalization without forced resizing, a YOLOv11 feature extractor, a multi-scale feature pyramid, a resolution-agnostic design in which the backbone and FPN accept unchanged image sizes and channels while dynamic anchor or objectness heads adapt to varying feature-map strides, and a YOLOv11-style instance-segmentation head followed by non-maximum suppression, thresholding of mask logits, and contour extraction into closed field boundaries (Lavreniuk et al., 3 Apr 2025). The summary states explicitly that no fixed spatial resampling is performed, so the network accepts $\Phi_C$8 of arbitrary $\Phi_C$9 (Lavreniuk et al., 3 Apr 2025).

The field-domain DelAnyFlow then applies a five-stage post-processing sequence (Lavreniuk et al., 17 Nov 2025). Quality masking excludes pixels flagged as nodata or outside an external land-cover mask such as water, forest, or urban areas, and generates a context mask by buffering excluded zones by 20 m (Lavreniuk et al., 17 Nov 2025). Morphological refinement operates per instance in descending area order: erode by a $DF \subseteq (P \cup D) \times (P \cup D)$00 structuring element, retain only the largest connected component, dilate by the same element, and discard instances smaller than 0.1 ha for high-resolution inputs or 0.5 ha for Sentinel-2 (Lavreniuk et al., 17 Nov 2025). Cross-tile unification merges overlapping detections across adjacent $DF \subseteq (P \cup D) \times (P \cup D)$01 tiles when

$DF \subseteq (P \cup D) \times (P \cup D)$02

satisfies IoU $DF \subseteq (P \cup D) \times (P \cup D)$03 and overlap area $DF \subseteq (P \cup D) \times (P \cup D)$04 (Lavreniuk et al., 17 Nov 2025). Mosaic assembly rasterizes all refined instances to a seamless grid with at most one field per pixel and removes slivers smaller than $DF \subseteq (P \cup D) \times (P \cup D)$05 (Lavreniuk et al., 17 Nov 2025). Vectorization and topology validation extract contours, ensure closed loops, enforce no overlaps or gaps, remove small polygons, reproject to EPSG:4326, and embed attributes including field ID, area, and tile provenance (Lavreniuk et al., 17 Nov 2025).

A plausible implication is that the field-domain meaning of “flow” is procedural rather than dynamical: it refers to the sequential movement from instance masks to topology-valid vector boundaries, not to optical flow.

6. Data regimes, quantitative performance, and deployment scale

A substantial part of the field-domain DelAnyFlow literature is the dataset regime supporting resolution agnosticism. FBIS-22M is described as the largest field-delineation benchmark to date, with 672,909 image patches and 22.9 million manually validated field masks, spanning resolutions from 0.25 m to 10 m and geographic coverage across Austria, France, Luxembourg, the Netherlands, Slovakia, Slovenia, Spain, Sweden, and Ukraine (Lavreniuk et al., 17 Nov 2025). The related DelAny summary gives 22,926,427 individual-field masks and a split of 636,784 training images and 36,125 test images, using COCO-style JSON with one “field” category, bounding boxes, and per-instance segmentation polygons or masks (Lavreniuk et al., 3 Apr 2025). The annotation protocol includes official LPIS parcels where available and expert manual delineation on high-resolution imagery elsewhere, followed by quality control to remove slivers, ensure closed polygons, and correct topological errors (Lavreniuk et al., 17 Nov 2025).

Training settings are reported with high specificity. DelAny and DelAny-S are trained from COCO-pretrained weights for 30 epochs with batch size 320, corresponding to 40 images per GPU across 8 NVIDIA H100s, learning rate $DF \subseteq (P \cup D) \times (P \cup D)$06 decayed by a factor of 0.1 at epochs 20 and 25, and augmentations including random horizontal and vertical flips, color jitter, mixup, copy-paste, and mosaic for the first 20 epochs (Lavreniuk et al., 17 Nov 2025). The DelAny summary separately specifies AdamW, initial learning rate $DF \subseteq (P \cup D) \times (P \cup D)$07, exponential decay, 30 epochs, and inference on one $DF \subseteq (P \cup D) \times (P \cup D)$08 crop at $DF \subseteq (P \cup D) \times (P \cup D)$09 ms for DelAny-S and $DF \subseteq (P \cup D) \times (P \cup D)$10 ms for DelAny (Lavreniuk et al., 3 Apr 2025).

The following table consolidates quantitative values stated in the supplied field-domain sources.

System / setting	Metric	Reported value
DelAny	[email protected]	0.720
DelAny	mAP@[0.5:0.95]	0.477
SAM2	[email protected]	0.382
SAM2	mAP@[0.5:0.95]	0.235
DelAny	Latency per 512×512 tile on A100	25 ms
DelAny-S	Latency per 512×512 tile on A100	16.8 ms
Ukraine deployment	Area processed	603,000 km²
Ukraine deployment	End-to-end runtime	5.4 hours
Ukraine, 5 m	Fields detected	3.75 million
Ukraine, 2.5 m	Fields detected	5.15 million
Sinergise Solutions	Fields detected	2.66 million
NASA Harvest (2023)	Fields detected	1.69 million

The sources characterize the gain over SAM2 as $DF \subseteq (P \cup D) \times (P \cup D)$11 in [email protected] and $DF \subseteq (P \cup D) \times (P \cup D)$12 in mAP@[0.5:0.95], with inference approximately $DF \subseteq (P \cup D) \times (P \cup D)$13 faster than SAM2 in one summary and approximately $DF \subseteq (P \cup D) \times (P \cup D)$14 faster in another, reflecting the different exact latency comparisons reported in the two supplied descriptions (Lavreniuk et al., 17 Nov 2025, Lavreniuk et al., 3 Apr 2025). For the smallholder size range $DF \subseteq (P \cup D) \times (P \cup D)$15–$DF \subseteq (P \cup D) \times (P \cup D)$16 ha, DelAnyFlow is reported to detect up to 70% more fields than Sinergise and over 200% more than NASA Harvest, while in the $DF \subseteq (P \cup D) \times (P \cup D)$17–$DF \subseteq (P \cup D) \times (P \cup D)$18 ha range DelAnyFlow and Sinergise perform comparably (Lavreniuk et al., 17 Nov 2025).

The deployment example for Ukraine states that processing 603,000 km² using 5 m Sentinel-2 composites on a single AMD Ryzen 9 9900X + RTX 5070 Ti workstation with 64 GB RAM took 5.4 hours end-to-end, excluding data download (Lavreniuk et al., 17 Nov 2025). Another description states that a complete field boundary layer for Ukraine was generated in under six hours on a single workstation (Lavreniuk et al., 17 Nov 2025). These values are consistent at the scale of the stated deployment.

7. Limitations, interpretations, and relation among the variants

The field-domain literature states several limitations directly. FBIS-22M is Europe-centric, and African and Southeast Asian smallholder systems are under-represented (Lavreniuk et al., 17 Nov 2025). Single-date optical reliance can induce spurious boundaries under clouds and phenology changes, and integration of SAR and temporal composites is identified as a priority (Lavreniuk et al., 17 Nov 2025). Over-aggregation is reported in sparsely bounded areas, where some very large fields larger than 1,000 ha are incorrectly merged; adjusting merging thresholds by agro-ecological zone is suggested as a possible mitigation (Lavreniuk et al., 17 Nov 2025).

The optical-flow literature, by contrast, frames the central limitation as fragmentation or vulnerability at motion boundaries and under occlusion when models rely on local photometric cues without object-level information (Zhou et al., 2023, Yuan et al., 2024). The SAM-enhanced FlowFormer formulation addresses this through frozen global object-aware features plus task-specific adaptation (Zhou et al., 2023), whereas the ARFlow-based unsupervised formulation addresses it through semantic augmentation, homography-based regional smoothness, and optional mask-feature aggregation (Yuan et al., 2024). These are conceptually related responses to the same failure mode, but they remain architecturally distinct.

Across all usages, a recurring pattern is the combination of a broad delineation prior with task-specific structure. In the UCD-derived notation, the prior is the unified syntax of control flow, data flow, boundaries, and timelines (Polkovnikov, 2016). In the optical-flow variants, the prior is SAM-derived object integrity combined with flow-specific decoding or losses (Zhou et al., 2023, Yuan et al., 2024). In field delineation, the prior is DelAny’s large-scale, multi-resolution instance segmentation combined with structured geometric post-processing and vector topology enforcement (Lavreniuk et al., 17 Nov 2025, Lavreniuk et al., 3 Apr 2025).

This suggests that “DelAnyFlow” is best treated as a family resemblance term. The common theme is delineation of coherent entities—process structure, object motion, or agricultural parcels—using a pipeline that couples global structural cues with domain-specific mechanisms for consistency, adaptation, and final representation.