PuTR-CouT: Counting-by-Tracking in Camera-Trap Image Sequences

Fagner Cunha; Juan G. Colonna; Eulanda M. dos Santos

cs.CVSep 4, 2026

PuTR-CouT: Counting-by-Tracking in Camera-Trap Image Sequences

Authors: Fagner Cunha, Juan G. Colonna, Eulanda M. dos Santos

Organizations: Federal University of Amazonas Manaus, Amazonas, Brazil

Abstract

Species identification in camera trap images has been widely studied, but key ecological modeling tasks such as species abundance or density estimation also require counting individual animals. However, the lack of counting labels in most datasets and low frame rates (typically ~1 frame per second) make sequence-level tracking and count estimation particularly challenging. In this work, we present PuTR-CouT, a counting-by-tracking framework built on a transformer-based learned association mechanism for sequence-level animal counting in camera trap images. To address the scarcity of annotated tracking data, we generate synthetic training data by exploiting structural priors, such as static backgrounds and short temporal bursts, to heuristically create pseudo-tracking labels in a weakly supervised manner. The resulting tracker associates detections across frames, using these tracks to estimate per-species counts. We also refine the MaxBoxCount heuristic used by the top solutions of the iWildCam 2021 challenge as a strong baseline, setting the highest score reported to date. When evaluated on the iWildCam 2021 benchmark, our framework PuTR-CouT delivers competitive counting results compared to the improved MaxBoxCount, with the added capability of multi-species predictions and track-level verification.

Figures & tables

Figure 1: The PuTR-Cout framework first detects bounding boxes using MegaDetectorV1000, connects frame-to-frame detections via a synthetic-trained Transformer tracker to form single-animal tracks, and finally classifies each track independently to output per-species animal counts for the sequence.

Figure 2: Illustration of the proposed synthetic sequence generation. Base-case tracks captured at the same location and within the same coarse timeslot can be merged to simulate a multi-individual sequence. Each cropped box is treated as a token by PuTR, removing the need to reconstruct composite images.

Strategy	BBoxes	Classifier	Re-ID	Counting	Classif.	Public $\downarrow$	Private $\downarrow$
2021 Winning solution [ 9 ]	MDv4	Eff.-B2	–	Boxes	Sequence	0.0300970	0.0293278
2nd place solution	MDv3	Eff.-B2	–	Boxes	Sequence	0.0297267	0.0293786
3rd place solution	MDv4	ResNet152 + Eff.-B3 + Eff.-B7	–	Boxes	Sequence	0.0315964	0.0308570
MaxBoxCount (ours)	MDv4	Swin-B	–	Boxes	Sequence	0.0269689	0.0256999
MaxBoxCount (ours)	MDv1000	Swin-B	–	Boxes	Sequence	0.0254706	0.0238785
ByteTrack	MDv1000	Swin-B	–	Tracks	Sequence	0.0297142	0.0282785

Table 1: Main results on iWildCam 2021 for animal counting (MCRMSE). We compare the top-3 challenge submissions, the improved MaxBoxCount baseline, classical multi-object trackers, and PuTR-CouT. All classifier entries are ensembles of the specified architecture. Bold and underlined values indicate the best and second-best scores, respectively. Lower is better.

Figure 3: Illustrative examples of counting predictions. Bounding box colors denote the tracks generated by PuTR-CouT. Only the first three images of each sequence are displayed. The first row demonstrates a key advantage of PuTR-CouT over MaxBoxCount in handling multi-species counting scenarios. Best viewed in color.

Tracker	Reorder		Filter	Private $\downarrow$
	Train	Test	BBoxes
ByteTrack				0.0282683
ByteTrack		$\checkmark$		0.0284988
BoT-SORT (MegaDesc.)				0.0270292
BoT-SORT (MegaDesc.)		$\checkmark$		0.0272918
PuTR (MegaDesc.)				0.0256394

Table 2: Effect of reordering camera-trap sequences based on EXIF information for the training and test sets, and pre-filtering empty bounding boxes. Results are reported on the private test set. Lower is better.

Tracker	IoU comp.	HMIOU	HIOU	Traj. conf.	Det. conf.	Private $\downarrow$
PuTR						0.0250820
PuTR	$\checkmark$					0.0252154
PuTR		$\checkmark$				0.0253234
PuTR			$\checkmark$			0.0253234
PuTR				$\checkmark$		0.0254082
PuTR					$\checkmark$	0.0251428

Table 3: Ablation of affinity-matrix components in PuTR on camera-trap sequences. Results are reported on the private test set. Lower is better.

Object Counting

Object counting estimates the number and locations of target instances in images or video, from dense crowds to fine-grained categories.

See all 61 papers

Counting Animals in Camera-Traps Image Sequences without Count Labels: Winning Solution to the iWildCam 2021 Challenge
Sep 3, 2026Fagner Cunha, Juan G. Colonna, Eulanda M. dos SantosObject Counting Wildlife Monitoring
Federal University of Amazonas · Manaus, Amazonas, Brazil
Multi-Object Tracking Consistently Improves Wildlife Inference
May 15, 2026Mufhumudzi Muthivhi, Jiahao Huo, Fredrik Gustafsson +1Temporal Consistency Wildlife Monitoring
Academy of Computer Science and Software Engineering University of Johannesburg Johannesburg, South Africa · Department of Electrical Engineering Linköping University Linköping, Sweden
Promptable Animal Pose Tracking Across Species
Aug 5, 2026Le Li, Daniela Ivanova, Nicolas PugeaultKeypoint Detection
School of Computing Science University of Glasgow

Sep 3, 2026J/K move · Enter open · S save

Counting Animals in Camera-Traps Image Sequences without Count Labels: Winning Solution to the iWildCam 2021 Challenge

Fagner Cunha, Juan G. Colonna, Eulanda M. dos Santos

Federal University of Amazonas · Manaus, Amazonas, Brazil

Object Counting Wildlife Monitoring

Open paper AskPDF arXiv

Camera traps have become an essential tool for wildlife monitoring, motivating the development of computer vision methods for the automated extraction of information from these data. While most prior work has focused on species identification, many ecological applications also require estimating the number of unique individuals appearing across short image sequences. This task is particularly challenging because camera traps typically acquire bursts of images at approximately one frame per second, creating large temporal discontinuities that may make conventional multi-object tracking methods unreliable, and because manually collecting individual count annotations is prohibitively expensive. In this work, we describe the winning solution to the iWildCam 2021 Challenge, which introduced a benchmark for counting animals at the sequence level under realistic annotation constraints where count annotations are unavailable for training. Our approach, MaxBoxCount, combines a strong species classification pipeline with a simple yet effective counting heuristic based on MegaDetector detections to estimate the number of unique individuals without requiring count annotations. Code is available at https://github.com/alcunha/iwildcam2021ufam.

Abstract

Object Counting

Explore similar work

Counting Animals in Camera-Traps Image Sequences without Count Labels: Winning Solution to the iWildCam 2021 Challenge

Multi-Object Tracking Consistently Improves Wildlife Inference

Promptable Animal Pose Tracking Across Species