Synthetic Evaluation Dataset for AI-Driven Traffic Violation Detection in Philippine Road Environments (CARLA, 175 clips)

ATVD Synthetic Dataset Title. Synthetic Evaluation Dataset for AI-Driven Traffic Violation Detection in Philippine Road Environments (CARLA, 175 clips) Author. Sheldon Arthur M. Sagrado, Department of Computer Science, University of the Philippines Cebu. Companion paper. Artificial Intelligence-Driven Traffic Violation Detection Using Multi-Object Tracking and Geometric Rules in Philippine Road Environments (December 2025). Abstract This deposit hosts the synthetic evaluation dataset for an Automated Traffic Violation Detection (ATVD) pipeline that couples a YOLO26l detector with BYTETrack multi-object tracking and calibrated geometric rules with dwell-time confirmation for five violation types: no-stopping, illegal U-turn, restricted lane, counterflow, and overspeed. The dataset is a 175-clip rollup generated in CARLA across five scenarios (counterflow, illegal_uturn, no_stopping, overspeed, restricted_lane) crossed with weather (clear, cloudy, rain), time of day (noon, sunset, night), and three variation packs (v1, v2, v3). Each clip ships with per-frame ground truth, the site configuration used by the pipeline, and the per-clip evaluation JSON produced by the evaluation harness. File inventory synthetic-clips/<scenario>/<clip>/ , 175 clip directories. For restricted_lane, clips are nested one level deeper under per-class subdirectories (bus/, motorcycle/, truck/). Each leaf directory contains: video.mp4 , rendered clip config.yaml , site configuration (regions, rule parameters, calibration) ground_truth.json , per-frame ground truth predictions.json , pipeline predictions (carried for reproducibility) evaluation.json , per-clip evaluation summary info.md , scenario notes ref_points.png , calibration reference image evaluation/ , rolled-up evaluation tables: synthetic_rollup.csv , one row per clip, 175 rows plus header synthetic_partial.csv , intermediate partial rollup (kept for traceability) real_rollup.csv , real-footage rollup (kept here to keep the evaluation tables co-located; the real-footage videos themselves are not part of this Zenodo deposit, see the paper's Data and Code Availability chapter for on-request access) checksums.sha256 , SHA-256 hash for every file in this deposit (excluding the checksum file itself). Verify any single file with: sha256sum -c checksums.sha256 --ignore-missing Per-scenario clip counts Scenario Clips counterflow 27 illegal_uturn 27 no_stopping 27 overspeed 27 restricted_lane 67 Total 175 Variation pack split: 49 v1, 63 v2, 63 v3. Total size Approximately 6.5 GB across 175 clips plus evaluation tables and checksums. License Released under the Creative Commons Attribution 4.0 International License (CC-BY-4.0). You are free to share and adapt the dataset with attribution. Citation @misc{sagrado2025atvd_synthetic, author = {Sheldon Arthur M. Sagrado}, title = {Synthetic Evaluation Dataset for AI-Driven Traffic Violation Detection in Philippine Road Environments (CARLA, 175 clips)}, year = {2025}, publisher = {Zenodo}, doi = {<Zenodo DOI placeholder>}, url = {<Zenodo record URL placeholder>} } Companion paper citation: @thesis{sagrado2025atvd, author = {Sheldon Arthur M. Sagrado}, title = {Artificial Intelligence-Driven Traffic Violation Detection Using Multi-Object Tracking and Geometric Rules in Philippine Road Environments}, school = {University of the Philippines Cebu}, year = {2025}, type = {Undergraduate Special Project} } Notes for users Six researcher-captured real-footage clips across five sites in Cebu are kept private and available on request to the corresponding author, subject to a confidentiality agreement aligned with the Data Privacy Act of 2012 (RA 10173). See the paper's Data and Code Availability chapter for the contact route. The implementation that produced these clips and evaluation tables is published as the ATVD system source repository linked from the same chapter.