Building in the open

AI-assisted video annotation tool with active learning propagation.

Auto-split video into segments, pre-annotate with YOLO/SAM, and propagate corrections across frames with active learning. Corrections auto-apply to past and future frames. Supports re-tracking of objects however small, in all directions, with classification updates.

Unreal Engine based synthetic video data generator for training and validation.

Procedurally generate millions of life-like traffic videos with perfect ground truth labels. Configurable environments, weather, lighting, vehicle density, and camera angles. Built on Unreal Engine for photorealistic rendering.

Multi-frame license plate super-resolution from video clips.

Fuses multiple video frames of a license plate bounding box to produce a single crisp, readable plate image. Based on BasicVSR++ / Real-ESRGAN / RVRT architectures. Solves the universal ANPR problem of blurry plates from surveillance cameras.

Benchmark suite for evaluating AI models on Indian traffic scenarios.

Standardized evaluation framework for detection, tracking, ANPR, and scene understanding in Indian traffic conditions. Covers edge cases like three-wheelers, overloaded trucks, no-lane driving, and mixed traffic.

Diverse Indian license plate dataset with varied formats, fonts, and languages.

Curated dataset of Indian license plates covering all state codes, bilingual text, fancy fonts, dealer plates, government plates, and degraded conditions (blur, night, rain). Annotations include plate text, bounding box, and metadata.

100 challenging Indian traffic scenes for model evaluation.

100 hand-picked, extremely challenging traffic scenes from Indian roads. Dense mixed traffic, unusual vehicles, poor visibility, complex intersections. Designed to stress-test detection and tracking models.

Vehicle tracking toolkit optimized for Indian traffic conditions.

Multi-object tracking toolkit with re-identification, occlusion handling, and camera handoff. Tuned for dense, chaotic traffic environments where standard trackers fail.

Detection model configs and weights for Indian traffic objects.

Pre-trained detection models and configuration files covering 50+ Indian vehicle types including auto-rickshaws, tempos, tractors, bullock carts, and more. Ready-to-deploy configs for TensorRT and ONNX Runtime.