Classification of natural indoor and outdoor scenes from radiometric, photometric, and colorimetric features
Bibliographic record
Abstract
Background: In natural, real-world scenes, environmental light can vary significantly in intensity, spectrum, color, and spatial and temporal characteristics due to the presence of different light sources (daylight, electric lighting, self-luminous displays, and mixtures thereof) and light-surface interactions. The ‘spectral diet’ an observer is exposed to reflects this complexity, compounded by body, head, and eye movements. Understanding some of this complexity requires a systematic and detailed understanding of the variation of light in the real world. From the first principles, we know that light outdoors is of higher intensity than light indoors. A detailed survey of the spectral, spatial, and temporal features of light in the real world, which is highly relevant for architectural design, occupational health, and environmental medicine, has thus far not been undertaken. Methods: In the SCENES Dataset (https://www.scenes-dataset.org/), we have comprehensively characterized the spectral, spatial, and temporal variations of natural scenes using a novel multimodal data collection setup comprising an α- opic imaging radiometer, a high-resolution spectroradiometer, illuminance and colorimetric measurement devices, a depth camera, and an uncalibrated wide- field RGB video camera. All instruments were integrated into a portable box for easy deployment with a power supply through external batteries. Data were collected across various times of day and seasons. Each scene was described using a novel metadata descriptor (n=43 items), encompassing detailed information, including geographical information, weather conditions, and scene categories (9 indoor subcategories, 11 outdoor subcategories). To understand the basic aspects of the datasets, we used descriptive statistics (mean, SD, min., max.), and applied the random forest algorithm to develop a scheme for indoor vs. outdoor specifications on spectrum-derived data. Results: We measured natural scenes both indoors (n=313) and outdoors (n=366) in five locations (Tübingen, Germany: n=53 indoors, n=254 outdoors; Munich, Germany: n=61 indoors, n=11 outdoors; Prague, Czech Republic: n=19 indoors, n=16 outdoors; Lyon, France: n=67 indoors, n=26 outdoors; Ottawa, Canada: n=113 indoors, n=59 outdoors). As expected, there are key differences between indoor vs. outdoor scenes in photopic illuminance (mean±SD 1826.51±7402.82 lx [min. 13.8 lx, max. 63721.1 lx] indoors vs. 13401.91±7402.82 lx [0.28 lx, 110872.8 lx] outdoors) and melanopic equivalent daylight illuminance (mean±SD 1586±6416.93 lx [min. 11.9 lx, max. 55178.3 lx] indoors vs. 12240.01±18741.81 lx [0.19 lx, 97807.52 lx] outdoors). The results from the random forest algorithm indicate excellent model performance (accuracy: 0.963, precision: 0.982, recall: 0.931, F1: 0.956). Regarding feature importance in the classification, CRI Ra (color rendering index) was ranked highest. Conclusion: In this study, we collected indoor and outdoor natural scene data across various geographical contexts. A preliminary analysis has evaluated the ability of several spectrum-derived features to allow for the classification of indoor vs. outdoor scenes. Further analyses will investigate the possibility of classifying scene subcategories and probe the spatiotemporal characteristics of natural scenes in further detail. The SCENES Dataset will be made available as an open-access dataset to serve as a benchmark of the properties of environmental light.
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 | 0.002 |
| Meta-epidemiology (narrow) | 0.003 | 0.003 |
| Meta-epidemiology (broad) | 0.004 | 0.001 |
| Bibliometrics | 0.013 | 0.039 |
| Science and technology studies | 0.000 | 0.002 |
| Scholarly communication | 0.001 | 0.001 |
| Open science | 0.002 | 0.001 |
| Research integrity | 0.005 | 0.004 |
| Insufficient payload (model declined to judge) | 0.002 | 0.002 |
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; both teacher heads agree on what is shown here.
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".