MOO: A Multi-view Oriented Observations Dataset for Viewpoint Analysis in Cattle Re-Identification

William Grolleau; Achraf Chaouch; Astrid Sabourin; Guillaume Lapouge; Catherine Achard

cs.CVMar 4, 2026

MOO: A Multi-view Oriented Observations Dataset for Viewpoint Analysis in Cattle Re-Identification

Authors: William Grolleau, Achraf Chaouch, Astrid Sabourin, Guillaume Lapouge, Catherine Achard

Organizations: Universite Paris-Saclay, CEA, List, F-91120 Palaiseau, France · Sorbonne University, CNRS, ISIR, 4 Place Jussieu, 75005 Paris, France

Abstract

Animal re-identification (ReID) faces critical challenges due to viewpoint variations, particularly in Aerial-Ground (AG-ReID) settings where models must match individuals across drastic elevation changes. However, existing datasets lack the precise angular annotations required to systematically analyze these geometric variations. To address this, we introduce the Multi-view Oriented Observation (MOO) dataset, a large-scale synthetic AG-ReID dataset of $1,000$ cattle individuals captured from $128$ uniformly sampled viewpoints ( $128,000$ annotated images). Using this controlled dataset, we quantify the influence of elevation and identify a critical elevation threshold, above which models generalize significantly better to unseen views. Finally, we validate the transferability to real-world applications in both zero-shot and supervised settings, demonstrating performance gains across four real-world cattle datasets and confirming that synthetic geometric priors effectively bridge the domain gap. Collectively, this dataset and analysis lay the foundation for future model development in cross-view animal ReID. MOO is publicly available at https://github.com/TurtleSmoke/MOO.

Figures & tables

Figure 1: MOO: A synthetic dataset for systematic analysis and real-world transferability. The workflow encompasses (1) individual generation; (2) dataset rendering; and (3) systematic analysis and real-world transfer.

Species	Dataset	Individuals	Images	View	Synthetic	Non-Correlated Images	Annotation
Wildlife	ATRW [ 19 ]	107	3,649	Lateral	$\circ$	$\bullet$
	SealID [ 24 ]	57	2,080	Lateral	$\circ$	$\bullet$
	SeaTurtleIDHeads [ 1 ]	438	8,729	Lateral	$\circ$	$\bullet$
	Nyala Data [ 11 ]	237	1,942	Lateral	$\circ$	$\bullet$	Left vs Right
	LeopardID [ 6 ]	430	6,805	Lateral	$\circ$	$\bullet$
	CzechLynx [ 31 ]	219+300	37k+100k	Lateral	$\bullet$	$\bullet$

Table 1: Comparison of wildlife and cattle re-identification datasets with MOO. Non-Correlated Images : independent frames reducing temporal and background bias. Synthetic : includes synthetic data. Annotation : viewpoint labels provided as continuous angles (Az./El.) or discrete categories.

Figure 2: mAP per training elevation range across eight partitions evaluated in a same-view setup, where Query and gallery images share the same elevation.

Figure 3: mAP for the best Single-View expert (black dot) and All-View model (blue curve) across eight partitions evaluated in a same-view setup, where Query and gallery images share the same elevation.

Panel of Table 2 — Table 2: mAP across four azimuth training and evaluation scenarios evaluated in a same-view setup, where Query and gallery images share the same view.

	Top $\rightarrow$ Side		Side $\rightarrow$ Top		Top-Side $\rightarrow$ Top-Side
Training	mAP	R-1	mAP	R-1	mAP	R-1
Train Side	13.1	35.6	13.0	75.5	20.6	97.4
Train Top	22.0	94.7	25.0	29.7	39.4	87.2
Train Top-Side	39.4	96.2	41.6	79.7	52.5	99.7

Table 3: mAP and Rank-1 comparison for three training configurations across three evaluation (query

\rightarrow

gallery) scenarios.

	FC15		FC17		AC17		C21
Initialization $\rightarrow$ Pre-training	mAP	R-1	mAP	R-1	mAP	R-1	mAP	R-1
Zero-shot (Direct Transfer)
ImageNet21K (Baseline)	51.1	46.8	45.8	60.0	55.5	79.6	9.4	32.4
ImageNet $\rightarrow$ MOO (All)	59.2	56.0	40.1	57.1	67.5	88.0	13.4	41.2
ImageNet $\rightarrow$ MOO (Top)	63.4	56.7	53.9	63.3	65.3	84.4	32.1	67.4
Supervised (Trained on Target)

Table 4: Transfer learning results on four real-world datasets. We compare the standard ImageNet baseline with our pre-training on MOO. Top: Zero-shot evaluation. Bottom: Supervised training on the target dataset.

Animal Re-Identification

Animal re-identification matches observations of the same individual across time, cameras, viewpoints, and changing conditions.

See all 33 papers

3D-LENS: A 3D Lifting-based Elevated Novel-view Synthesis method for Single-View Aerial-Ground Re-Identification
Apr 29, 2026William Grolleau, Astrid Sabourin, Guillaume Lapouge +1Novel View Synthesis Person Re-Identification
Université Paris-Saclay, CEA, List, F-91120 Palaiseau, France · Sorbonne University, CNRS, Institute of Intelligent Systems and Robotics (ISIR), 4 Place Jussieu, 75005 Paris, France
3D-VITAL: Visibility-aware Identity Training from Artificial Liftings of 3D Reconstruction model for Cross-View Object Re-Identification
Oct 12, 2026William Grolleau, Astrid Sabourin, Guillaume Lapouge +1Person Re-Identification
Université Paris-Saclay, CEA, List, F-91120 Palaiseau, France · Sorbonne University, CNRS, Institute of Intelligent Systems and Robotics (ISIR), Paris, France
Animal Re-Identification on Microcontrollers
Dec 9, 2025Yubo Chen, Di Zhao, Yun Sing Koh +1Efficient Neural Network Inference Animal Re-Identification
University of Auckland Auckland, New Zealand

Apr 29, 2026J/K move · Enter open · S save

3D-LENS: A 3D Lifting-based Elevated Novel-view Synthesis method for Single-View Aerial-Ground Re-Identification

William Grolleau, Astrid Sabourin, Guillaume Lapouge, Catherine Achard

Université Paris-Saclay, CEA, List, F-91120 Palaiseau, France · Sorbonne University, CNRS, Institute of Intelligent Systems and Robotics (ISIR), 4 Place Jussieu, 75005 Paris, France

Novel View Synthesis Person Re-Identification

Open paper AskPDF arXiv

Aerial-Ground Re-Identification (AG-ReID) is constrained by the viewpoint-domain gap, as drastic viewpoint disparities occlude or distort discriminative features, making cross-viewpoint image retrieval challenging. While existing methods rely on paired cross-view annotations, real-world deployments, such as wilderness search-and-rescue (SAR), often lack target-domain data, requiring retrieval from ground-level references alone. To our knowledge, we are the first to address this challenge by formalizing the Single-View AG-ReID (SV AG-ReID) setting, where models trained on a single real viewpoint must generalize to an unseen viewpoint. We propose 3D Lifting-based Elevated Novel-view Synthesis (3D-LENS), a unified framework combining geometrically-consistent novel view synthesis that leverages large-scale 3D mesh reconstruction, with a robust representation learning scheme to mitigate synthetic-to-real bias. Unlike 2D generative baselines that suffer from geometric inconsistencies or prior 3D methods that are restricted to class-specific templates, our approach ensures view-consistent synthesis across diverse categories without predefined templates that fail to capture fine-grained details, such as carried objects. Extensive experiments demonstrate that our method achieves state-of-the-art performance on SV AG-ReID scenarios. Code and data will be released at https://github.com/TurtleSmoke/3D-LENS.

Abstract

Animal Re-Identification

Explore similar work

3D-LENS: A 3D Lifting-based Elevated Novel-view Synthesis method for Single-View Aerial-Ground Re-Identification

3D-VITAL: Visibility-aware Identity Training from Artificial Liftings of 3D Reconstruction model for Cross-View Object Re-Identification

Animal Re-Identification on Microcontrollers