Accurate detection and counting of Gobiocypris rarus larvae are essential for laboratory toxicological experiments and aquatic biological research. However, the larvae detection task remains challenging due to small object size, dense aggregation, rapid movement, and complex imaging backgrounds. To address these challenges, this study proposes an improved detection framework termed YOLOv8-SLG, an optimized small-object detection framework built upon YOLOv8n and specifically designed for high-density fish larvae scenarios. The proposed model incorporates three task-oriented structural enhancements. First, a lightweight C3Ghost module is introduced into the Neck to reduce parameter redundancy while maintaining multi-scale feature fusion capability. Second, an enhanced SPPF-LSKA module integrates large-kernel spatial attention into multi-scale feature aggregation, strengthening contextual modeling and boundary discrimination. Third, an SPD-Conv module preserves fine-grained spatial information during downsampling, mitigating detail loss that typically degrades small-object localization performance. Experimental results on a self-constructed laboratory dataset demonstrate consistent improvements over the baseline YOLOv8n. YOLOv8-SLG increases Precision by 2.6 percentage points and Recall by 1.9 percentage points, mAP@0.5 by 0.7%, and notably mAP@0.5–0.95 by 3.4%. The larger gain under stricter IoU thresholds indicates improved localization performance under the evaluated conditions. Despite a moderate increase in computational complexity (from 8.1 to 11.6 GFLOPs), YOLOv8-SLG maintains a compact model size (6.6 MB), indicating an accuracy–computation trade-off for improved detection performance under controlled laboratory conditions. Overall, YOLOv8-SLG provides improved detection accuracy, with an acceptable increase in computational cost under controlled laboratory conditions, and interpretability, providing a promising and reproducible framework for automated larvae detection and counting under controlled laboratory conditions.
Low-field magnetic resonance imaging includes portable and open architectures, but quantitative results remain sensitive to field strength, temperature, environment, system state, acquisition, reconstruction, and model fitting. This methodological review asks what evidence supports claims of repeatability, intermediate precision, reproducibility, qualified reference-linked bias, measurement uncertainty, and use relevance. We conducted a targeted synthesis of empirical low-field reports, quantitative-imaging methods, metrology guidance, and MRI standards using a documented search and explicit source-classification criteria. The resulting framework separates three independent axes: represented precision conditions, strength of reference linkage, and relevance to the intended technical or clinical use. Each claim is linked to an experimental design, statistical estimand, configuration record, uncertainty model, and decision rule. An explicitly synthetic low-field T1 example illustrates design identifiability, precision-based replication planning, mixed-effects analysis, an uncertainty budget, and a guard-banded technical decision; it is not experimental validation. A staged pathway allows laboratories to begin with provenance, constancy, and within-system repeatability before progressing to reference-linked or multicentre studies. Meeting a phantom threshold establishes only the specified technical property under the tested conditions; it does not establish diagnostic utility, clinical benefit, regulatory conformity, or CE marking. The framework is an operational synthesis, not a replacement standard or validated quality score.
Unsupervised clustering of deep image features is often proposed for unlabeled image archives, yet it is usually judged by internal criteria rather than by agreement with expert annotations. In a single domain of texture images, namely full-size rock-core scans, we evaluate unsupervised partitions against depth-registered expert descriptions of 20 wells (nine classes), holding out entire wells. Color statistics, texture descriptors, VGG16, two DINOv2 models and SimCLR encoders pre-trained on the images are combined with k-means, Gaussian mixtures and SCAN, several seeds, and pre-training runs. Across all combinations, the mean class accuracy plateaus at 0.35–0.40 (chance level 0.12), and six color statistics reach this plateau with a Gaussian mixture model (0.37). Apparent gains of in-domain self-supervised features depend on the clustering algorithm and the pre-training run. The plateau is robust to the evaluation protocol, a coarser class scheme and more diverse fitting wells, and persists on color-normalized and grayscale images. Linear probes on the same frozen features reach 0.47–0.54, which locates the bottleneck in the unsupervised objectives. Fitting the clustering transductively on all unlabeled images lifts in-domain features to 0.42–0.48, but this gain does not transfer to unseen wells. The conclusions are specific to this domain.
Breast dynamic contrast-enhanced MRI acquires one image volume before contrast injection and several after it; each volume is a contrast phase. Which phases a tumour segmentation model needs has not been systematically tested. In 980 examinations from 22 sites of ISPY2, a neoadjuvant trial cohort in the public MAMA-MIA benchmark, we trained a 3D U-Net on all fifteen combinations of the first four phases, five seeds each, and measured overlap with the expert-corrected reference in Dice points (0–100). One comparison was pre-registered as confirmatory. The confirmatory comparison was inconclusive: a pair chosen before the sweep by a phase-selection model exceeded the clinical pre-contrast plus first post-contrast pair by 1.33 points (−0.39 to 3.12), which is too small a difference for this study to detect. Replacing the clinical pair with the best pair on average was also inconclusive (+1.56). All further findings are exploratory. Averaged across the three pairs of each kind, post-contrast pairs exceeded pairs containing the pre-contrast volume by 2.52 points (1.26–3.91), but this concerned overlap, not boundary accuracy, was not significant in a second cohort, reversed in point estimate when the ISPY2 models were applied there without retraining, and could be inflated by how the reference segmentations were generated. No pair suited every patient: the clinical pair was best for 30% of patients and the best-on-average pair for only 12%, so the average does not identify the best pair for an individual. The interval from pre-contrast to first post-contrast volume varied 2.25-fold across sites (87–196 s), so phase index cannot be assumed to mark a fixed point on the enhancement curve. For data already acquired, phase choice is cost-free, but its effect is small, patient-specific and unconfirmed beyond one cohort; it does not warrant protocol changes. Recording injection time would allow phase timing to be compared across sites.
Square fiducial markers provide a robust and efficient way to establish visual references in camera localization tasks, combining reliable detection with unique identification through an embedded binary code. The robustness of these systems depends largely on the minimum distance between valid codes: larger distances enable the correction of more bit errors and reduce the risk of marker misidentification. This paper presents two approaches for generating square-marker dictionaries with large inter-marker distances. The first is an exact approach that formulates the dictionary-generation problem as a maximum-clique problem. It finds optimal solutions for a broader range of cases than the previous exact method, although its memory and computational requirements still restrict its practical applicability to small marker sizes (3×3 and 4×4 bits). The second is a packing-based suboptimal approach that constructs the dictionary incrementally and scales to larger code and dictionary sizes. Experimental results demonstrate that the proposed approach improves upon existing dictionary-generation methods, enabling the construction of larger dictionaries with enhanced error-detection and error-correction capabilities. The source code and generated dictionaries are publicly available.
