How pranjal1-1/urbanpulse-ai works
UrbanPulse AI is currently a RoadVision dataset-preparation, training, and evaluation foundation—not the full bus-sensing and GIS platform described as its longer-term pipeline. The main implemented workflow converts RDD2022 annotations into a prepared dataset, resolves and validates that dataset and training configuration, runs training, then evaluates a checkpoint and reports support-aware metrics. Contracts and Kaggle experiment tooling are additional branches. Future edge perception, multi-bus fusion, and authority-action stages are not represented as implemented components.
Main entry points: train.py Training orchestration · evaluate.py Checkpoint evaluator · rdd2022.py RDD2022 converter · prepare_rdd2022.py Preparation CLI
Components 22
Data preparation
- Preparation CLI
scripts/prepare_rdd2022.py - RDD2022 converter
ai/data/rdd2022.py - Prepared dataset
AI contracts
- Dataset contract
ai/contracts/dataset.py - Canonical taxonomy
ai/contracts/taxonomy.py - Event schema
ai/contracts/events.py - AI data config
configs/datasets/SIH26124_AI_Data_Config_v1.json
RoadVision lifecycle
- RoadVision config
ai/roadvision/config.py - Dataset resolver
ai/roadvision/dataset.py - Training CLI
scripts/train_roadvision.py - Training orchestration
ai/roadvision/train.py - Runtime capture
ai/roadvision/environment.py - Run provenance
ai/roadvision/manifest.py - Model checkpoint
- Evaluation CLI
scripts/evaluate_roadvision.py - Checkpoint evaluator
ai/roadvision/evaluate.py - Metric reporting
ai/roadvision/metrics.py
Experiments
- Kaggle runner
experiments/kaggle/kaggle_runner.py - Speed experiment
experiments/kaggle/speed_experiment.py - Speed telemetry
experiments/kaggle/speed_telemetry.py
Other components
- Engineer
- RDD2022 source
Connections 24
- Engineer → Preparation CLI: runs preparation
README.md - Preparation CLI → RDD2022 converter: calls converter
scripts/prepare_rdd2022.py - RDD2022 source → RDD2022 converter: reads annotations
ai/data/rdd2022.py - RDD2022 converter → Prepared dataset: writes dataset
ai/data/rdd2022.py - Preparation CLI → Dataset contract: validates contract
scripts/prepare_rdd2022.py - RDD2022 converter → Canonical taxonomy: uses labels
ai/data/rdd2022.py - Canonical taxonomy → AI data config: loads taxonomy
ai/contracts/taxonomy.py - Event schema → Canonical taxonomy: validates classes
ai/contracts/events.py - Engineer → Training CLI: runs training
README.md - Training CLI → RoadVision config: loads settings
scripts/train_roadvision.py - Training CLI → Training orchestration: invokes training
scripts/train_roadvision.py - Training orchestration → Dataset resolver: resolves dataset
ai/roadvision/train.py - Dataset resolver → Prepared dataset: reads dataset
ai/roadvision/dataset.py - Training orchestration → Runtime capture: captures runtime
ai/roadvision/train.py - Training orchestration → Run provenance: records run
ai/roadvision/train.py - Training orchestration → Model checkpoint: writes weights
ai/roadvision/train.py - Engineer → Evaluation CLI: runs evaluation
README.md - Evaluation CLI → Checkpoint evaluator: calls evaluator
scripts/evaluate_roadvision.py - Checkpoint evaluator → Model checkpoint: loads weights
ai/roadvision/evaluate.py - Checkpoint evaluator → Dataset resolver: resolves dataset
ai/roadvision/evaluate.py - Checkpoint evaluator → Metric reporting: builds report
ai/roadvision/evaluate.py - Engineer → Speed experiment: runs experiment
- Speed experiment → Kaggle runner: uses runner
experiments/kaggle/speed_experiment.py - Speed experiment → Speed telemetry: uses telemetry
experiments/kaggle/speed_experiment.py