This repository contains processed data and analysis code accompanying the Camera2Carbon study.
.
├── README.md
├── LICENSE-CODE
├── LICENSE-DATA
├── code/
│ ├── README.md
│ ├── requirements.txt
│ ├── Figure1.ipynb
│ ├── Figure2.ipynb
│ ├── Figure3.ipynb
│ ├── Figure4.ipynb
│ ├── average_speed_model.py
│ ├── powerlaw_analysis.py
│ └── gwr_mgwr_analysis.py
└── data/
├── *.csv
├── *.pkl
└── simulation_results/*.txt
The release contains the processed numerical results and analysis workflows used for:
- reproducing Figures 1--4;
- calculating average-speed CO2 emission factors;
- fitting heavy-tail distributions to emission patterns; and
- fitting geographically weighted regression (GWR/MGWR) models for spatial associations with urban covariates.
The upstream cross-camera association and trajectory-recovery method is available in tsinghua-fib-lab/Cam-Traj-Rec.
| Path | Description | Related analysis |
|---|---|---|
data/jn_spatial.pkl |
Processed spatial emission and road-network quantities | Spatial patterns |
data/sz_spatial.pkl |
Processed spatial emission and road-network quantities for Shenzhen | Spatial patterns |
data/gwr_effects.pkl |
Selected geographically weighted regression effects | GWR summaries |
data/bytime_byvehicle_jn_w.csv |
Working-day aggregates by time window and vehicle class | Temporal and fleet composition |
data/bytime_byvehicle_jn_nw.csv |
Non-working-day aggregates by time window and vehicle class | Temporal and fleet composition |
data/grid_congestion_jn1.pkl |
Gridded congestion and emission quantities for Jinan dataset 1 | Congestion response |
data/grid_congestion_jn2.pkl |
Gridded congestion and emission quantities for Jinan dataset 2 | Congestion response |
data/jn_road_congestion_speed.pkl |
Road-level congestion, speed, and emission summaries | Congestion response |
data/bus_vs_private.pkl |
Processed public-transit/private-vehicle comparison | Per-passenger comparison |
data/reduction_spatial.pkl |
Spatial outputs for selected counterfactual scenarios | Scenario analysis |
data/depeak.pkl |
Original and temporally redistributed emission profiles | Departure-time scenario |
data/simulation_results/*.txt |
Scalar total-emission outputs from production simulation runs | Scenario-response analysis |
The CSV columns are start_window, VehicleType, Emission, Uncertainty,
RoadLength, and TravelTime. VehicleType is a categorical class rather than
a persistent vehicle identifier.
The pickle files contain Python objects used by the figure notebooks and require compatible versions of Python, NumPy, pandas, and NetworkX.
See code/README.md for installation and usage. The
Figure notebooks reproduce the main figure panels from the included processed
data. The standalone scripts provide the average-speed, heavy-tail, and
GWR/MGWR analysis workflows.
Files under data/simulation_results/ are outputs of the IM model. Each text file contains one aggregate emission result.
The filename suffix identifies the intervention and its parameter value:
| Suffix | Manuscript scenario | Meaning |
|---|---|---|
| no suffix | Baseline | Original simulation without an intervention |
s<rate> |
P1: Global Flow Restriction | All vehicle classes are randomly sampled at the specified retention rate |
e<rate> |
P2: Global Electrification Transition | The specified proportion of fuel-powered vehicles is converted to electric vehicles (electrify_rate) |
c<rate> |
P3: Modal Shift / Driving Reduction | Only private cars (types=[3,4], sedans and SUVs) are randomly sampled at the specified retention rate; other vehicle classes are retained |
t<rate> |
P4: Heavy Asset Restriction | Only heavy trucks (types=[1]) are randomly sampled at the specified retention rate; other vehicle classes are retained |
For the sampling scenarios (s, c, and t), rate is the fraction retained
in the simulator and the restricted/removed fraction is 1 - rate. For example,
s09 retains 90% of all vehicles, whereas c07 retains 70% of private cars.
The compact numeric encoding omits the decimal point: 005, 01, 095, and
10 denote 0.05, 0.10, 0.95, and 1.00, respectively.
P5 (Temporal Load Balancing / Departure Time Shifts) does not use an s, e,
c, or t filename suffix. Its processed original and redistributed temporal
profiles are stored in data/depeak.pkl. In this scenario, departure times are
shifted by no more than 30 minutes to reduce peak congestion without removing
travel demand.
If you use the Camera2Carbon data or analysis code, please cite the accompanying article:
Zhang, Y., Yu, F., Li, S., Li, M., Guan, D. & Li, Y. AI-enhanced sensing of instantaneous vehicular carbon flux via city-scale camera networks. Manuscript accompanying this repository.
@misc{zhang_camera2carbon,
author = {Zhang, Yunke and Yu, Fudan and Li, Sibo and Li, Meng and Guan, Dabo and Li, Yong},
title = {AI-enhanced sensing of instantaneous vehicular carbon flux via city-scale camera networks},
year = {2026},
note = {Manuscript accompanying the Camera2Carbon data and analysis-code repository},
url = {https://github.com/zyk21/C2C}
}If you use the upstream trajectory-recovery implementation, please also cite:
Yu, F., Ao, W., Yan, H., Zhang, G., Wu, W. & Li, Y. Spatio-Temporal Vehicle Trajectory Recovery on Road Network Based on Traffic Camera Video Data. Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining, 4413--4421 (2022). https://doi.org/10.1145/3534678.3539186
@inproceedings{yu2022spatiotemporal,
author = {Yu, Fudan and Ao, Wenxuan and Yan, Huan and Zhang, Guozhen and Wu, Wei and Li, Yong},
title = {Spatio-Temporal Vehicle Trajectory Recovery on Road Network Based on Traffic Camera Video Data},
year = {2022},
publisher = {Association for Computing Machinery},
doi = {10.1145/3534678.3539186},
booktitle = {Proceedings of the 28th ACM SIGKDD Conference on Knowledge Discovery and Data Mining},
pages = {4413--4421}
}Unless otherwise noted:
- source code under
code/is licensed under the MIT License; - processed datasets and aggregate simulation outputs under
data/are licensed under the Creative Commons Attribution 4.0 International License; - third-party software, data, and linked repositories remain subject to their original licenses and terms.
Copyright (c) 2026 Camera2Carbon authors.