cAIge Salmon Re-Identification Dataset, Segmentation Annotations, and Evaluation Results

The dataset is associated with the paper:

Høgstedt, E. U., Schellewald, C., Stahl, A., and Mester, R., Patch Ensembles for Robust Salmon Re-Identification with Weak Trajectory Labels, 2026 IEEE International Conference on Image Processing (ICIP), 2026.

This dataset accompanies the salmon re-identification framework presented in the repository:

salmon-reid-patch-ensemble GitHub repository

The upload contains datasets, segmentation annotations, trained model outputs, evaluation results, and cross-camera matching information used for salmon re-identification research in aquaculture environments.

The data was collected within the cAIge project and supports development of computer vision and AI methods for long-term monitoring of salmon in sea cages.

Contents

reid_dataset.zip

Contains the salmon re-identification dataset.
Folders 1–14 contain training data, folder 15 contains validation data, and folder 16 contains test data recorded from another camera for cross-camera evaluation.

segmentation.zip

Contains manually annotated segmentation data in LabelMe format, including annotations of Q1, Q2, and operculum regions.

analysis1.zip – analysis8.zip

Contain analysis data, embeddings, trajectories, intermediate evaluation outputs, and/or experiment results used in the associated re-identification analyses.

test.zip

Contains cross-camera test results for the main experiments, including ViT baseline and sliced-patch models.

a15_a16_idmatch_with_traj_IDs.xlsx

Contains manually verified identity matches between analysis folders 15 and 16 for cross-camera evaluation.

ap_per_query_all_models.zip

Contains per-query Average Precision (AP) values for all evaluated models, used for bootstrap statistical analysis.

Intended use

The dataset is intended for research in:

• fish re-identification
• aquaculture monitoring
• underwater computer vision
• fish tracking
• AI-based welfare assessment