Investigation of Deep Learning Optimization Algorithms in Scene Text Detection
Bibliographic record
Abstract
Scene text detection frameworks heavily rely on optimization methods for their successful operation. Choosing an appropriate optimizer is essential to performing recent scene text detection models. However, recent deep learning methods often employ various optimization algorithms and loss functions without explicitly explaining their selections. This paper presents a segmentation-based text detection pipeline capable of handling arbitrary-shaped text instances in wild images. We explore the effectiveness of well-known deep-learning optimizers to enhance the pipeline's capabilities. Additionally, we introduce a novel Segmentation-based Attention Module (SAM) that enables the model to capture long-range dependencies of multi-scale feature maps and focus more accurately on regions likely to contain text instances.The performance of the proposed architecture is extensively evaluated through ablation experiments, exploring the impact of different optimization algorithms and the introduced SAM block. Furthermore, we compare the final model against state-of-the-art scene text detection techniques on three publicly available benchmark datasets, namely ICDAR15, MSRA-TD500, and Total-Text. Our experimental results demonstrate that the focal loss combined with the Stochastic Gradient Descent (SGD) + Momentum optimizer with poly learning-rate policy achieves a more robust and generalized detection performance than other optimization strategies. Moreover, our utilized architecture, empowered by the proposed SAM block, significantly enhances the overall detection performance, achieving competitive H-mean detection scores while maintaining superior efficiency in terms of Frames Per Second (FPS) compared to recent techniques. Our findings shed light on the importance of selecting appropriate optimization strategies and demonstrate the effectiveness of our proposed Segmentation-based Attention Module in scene text detection tasks.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.002 | 0.004 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.001 | 0.004 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.001 | 0.000 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one teacher head, not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".