Evaluating Semantic Segmentation–Based Scene Descriptions for Multilane Rural Highway Point Clouds against Ground Truth
Bibliographic record
Abstract
This study investigates the use of two transformer-based models for generating semantic scene descriptions from semantically labeled point clouds of multilane rural highway segments, focusing on automating the extraction of detailed infrastructure elements. Two models were employed: Point Transformer v2, and a transformer-based point classification model with self-attention and cross-attention mechanisms, classifying 11 infrastructure classes along three rural highways in Alberta, Canada, capturing detailed point clouds under clear weather conditions. Each model incorporated specialized loss functions to address class imbalances and improve segmentation accuracy through a series of advanced neural network architectures, including self-attention and cross-attention layers to enhance inter-point relationships and model generalization. Two large language models (LLMs), Google Gemini and GPT-4o, were used to generate descriptive text based on the segmented point clouds. The LLMs provided detailed, structured narratives capturing visible highway features. The transformer-based point classification model outperformed Point Transformer v2, achieving an average F1 score of 71%, mean Intersection over Union (mIoU) of 61%, and precision exceeding 90% for critical infrastructure classes such as traffic signs, concrete barriers, light poles, and vegetation. Despite semantic segmentation accuracy differences, both models generated natural language scene descriptions that aligned closely with ground truth, with semantic similarity scores surpassing 80%. This consistency in descriptive accuracy underscores the robustness of both models in producing reliable, contextually relevant text. These results highlight the potential of integrating transformer-based models and LLMs for effective semantic segmentation and text description in rural highway infrastructure.
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 machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.001 | 0.004 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.001 |
| Bibliometrics | 0.003 | 0.001 |
| Science and technology studies | 0.000 | 0.001 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.001 | 0.001 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.002 | 0.001 |
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 source (direct Gemma or distilled Codex), 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".