Jerod Weinman

Leeje Jang, Yijun Lin, Yao-Yi Chiang et al.

Scene text spotting aims to detect and recognize text in real-world images, where instances are often short, fragmented, or visually ambiguous. Existing methods primarily rely on visual cues and implicitly capture local character dependencies, but they overlook the benefits of external linguistic knowledge. Prior attempts to integrate language models either adapt language modeling objectives without external knowledge or apply pretrained models that are misaligned with the word-level granularity of scene text. We propose TiCLS, an end-to-end text spotter that explicitly incorporates external linguistic knowledge from a character-level pretrained language model. TiCLS introduces a linguistic decoder that fuses visual and linguistic features, yet can be initialized by a pretrained language model, enabling robust recognition of ambiguous or fragmented text. Experiments on ICDAR 2015 and Total-Text demonstrate that TiCLS achieves state-of-the-art performance, validating the effectiveness of PLM-guided linguistic integration for scene text spotting.

Yijun Lin, Rhett Olson, Junhan Wu et al.

Text on historical maps provides valuable information for studies in history, economics, geography, and other related fields. Unlike structured or semi-structured documents, text on maps varies significantly in orientation, reading order, shape, and placement. Many modern methods can detect and transcribe text regions, but they struggle to effectively ``link'' the recognized text fragments, e.g., determining a multi-word place name. Existing layout analysis methods model word relationships to improve text understanding in structured documents, but they primarily rely on linguistic features and neglect geometric information, which is essential for handling map text. To address these challenges, we propose LIGHT, a novel multi-modal approach that integrates linguistic, image, and geometric features for linking text on historical maps. In particular, LIGHT includes a geometry-aware embedding module that encodes the polygonal coordinates of text regions to capture polygon shapes and their relative spatial positions on an image. LIGHT unifies this geometric information with the visual and linguistic token embeddings from LayoutLMv3, a pretrained layout analysis model. LIGHT uses the cross-modal information to predict the reading-order successor of each text instance directly with a bi-directional learning strategy that enhances sequence robustness. Experimental results show that LIGHT outperforms existing methods on the ICDAR 2024/2025 MapText Competition data, demonstrating the effectiveness of multi-modal learning for historical map text linking.

Chen Ziwen, Zixuan Guo, Jerod Weinman

Given stereo or egomotion image pairs, a popular and successful method for unsupervised learning of monocular depth estimation is to measure the quality of image reconstructions resulting from the learned depth predictions. Continued research has improved the overall approach in recent years, yet the common framework still suffers from several important limitations, particularly when dealing with points occluded after transformation to a novel viewpoint. While prior work has addressed this problem heuristically, this paper introduces a z-buffering algorithm that correctly and efficiently handles occluded points. Because our algorithm is implemented with operators typical of machine learning libraries, it can be incorporated into any existing unsupervised depth learning framework with automatic support for differentiation. Additionally, because points having negative depth after transformation often signify erroneously shallow depth predictions, we introduce a loss function to penalize this undesirable behavior explicitly. Experimental results on the KITTI data set show that the z-buffer and negative depth loss both improve the performance of a state of the art depth-prediction network.