manuelvaraletai/compaTAI-CDMX-Alpha

---
license: cc-by-nc-4.0
language:
- en
- es
pretty_name: compatai-CDMX-Alpha
size_categories:
- 100M<n<1B
task_categories:
- video-classification
- object-detection
- image-segmentation
- robotics
tags:
- mexico
- cdmx
- latin-america
- autonomous-driving
- traffic-analysis
- smart-cities
- edge-cases
- jsonl
- label-studio
- real-world-data
- chaos-detection
- behavioral-analysis
- pedestrian-safety
---

# 🚀 compaTAI: Mexico City Chaos Dataset (Video Alpha Sample)"

## Training AI to survive where the rules are suggestions.
"compaTAI presents a high-entropy Video Tracking Dataset captured in the chaotic heart of Mexico City (CDMX).
While standard datasets (Waymo, nuScenes) are built on predictable environments, compaTAI focuses on "The Chaos Edge Case":
the informal and non-linear traffic behaviors unique to Latin American megacities.

## 🌟 Why Video Tracking?
Unlike static image datasets, our video sequence provides temporal consistency.
This allows models to train for Multi-Object Tracking (MOT) and Behavior Prediction, essential for navigating environments
where lane markings don't exist and movement is erratic.

## 📊 Dataset SpecificationsTotal Duration: 52.06 Seconds.
* Total Frames: 1,562 (at 30 FPS).
* Resolution: High Definition (Processed for Fast-Start Streaming).
* Annotation Format: Label Studio JSONL (Temporal Bounding Box Sequences).
* Location 1: Mexico City (Critical Junctions).
* Location 2: State of Mexico (Critical Junctions)
* Type: 2D Video Rectangle Tracking with Interpolated Keyframes.

## 🏷️ The "Chaos" Taxonomy
Our labels capture the specific "Mexican Edge Cases" that standard sensors often misinterpret:

| Label | Description | Why it matters |
| ---------------------| -------------------------------------------------- | ------------------------------------------------ |
|Pedestrian_Irregular | Pedestrians crossing mid-avenue or between cars | Predicts non-linear human trajectory. |
|Street_Vendor | Mobile vendors navigating active traffic lanes. | Unique obstacle detection for informal economies.|
|Infrastructure_Deficit| Missing signals, potholes, or zero lane markings. | Trains defensive driving and path planning. |
|Moto_Filtering | Motorcycles weaving between lanes at high speed. | High-frequency proximity detection. |
|Microbus_Stop | Public transport stopping in the middle of the road.| Predicts sudden traffic flow interruptions. |

## 🛠️ Data Structure
The .jsonl file contains the temporal sequence of each object.
Each annotation includes a sequence array tracking:frame: The exact frame number.x, y, width, height: Normalized coordinates (0-100) for resolution-independent training.time: Exact timestamp within the 52-second clip.

## 🚀 How to AccessYou can download the raw video and the JSONL metadata directly from this repository to start training your trajectory prediction models.
Python# Coming soon: compaTAI utility script to visualize trackingfrom datasets import load_dataset
dataset = load_dataset("manuelvarale/compaTAI-CDMX-Chaos-Alpha")

## 💰 Commercial Version & Custom ServicesThis repository is a technical sample.
compaTAI offers specialized data for enterprise-grade autonomous systems:Full Video Datasets: 100+ hours of labeled CDMX traffic footage.
Specific Edge Cases: Custom recordings of "Rainy Nights" or "Peak Hour at Indios Verdes".
Custom Labeling: We use the compaTAI pipeline to label your raw data with our specialized taxonomy.
Enterprise Licensing & Custom Data Requests: [[e-mail]]

🔗 Visit our Hub: compatai.mx