/dhaka-ai

My solution for Dhaka Ai Traffic Detection Challenge.

Primary LanguageJupyter Notebook

Dhaka AI

My solution of Dhaka Ai Traffic Detection Challenge.

An international AI-based Dhaka Traffic Detection Challenge funded by Elsevier would be co-organized during STI 2020

Dataset

The dataset is composed of vehicle images, where an image contains a vehicle of one or more of 21 different classes of vehicle. This makes the dataset useful for multiple vehicle detection and recognition. The considered vehicle classes are: ambulance, auto-rickshaw, bicycle, bus, car, garbage van, human hauler, minibus, minivan, motorbike, Pickup, army vehicle, police car, rickshaw, scooter, Suv, taxi, three-wheelers (CNG), truck, van, wheelbarrow.

Inference on test image

Inference on test image Inference on test image

Train data overview

Train Data Overview

Train results

Train Result

Validation batch predictions

test_batch[0-9]+_pred.jpg shows—

validation batch predictions validation batch predictions

Validation batch labels

test_batch[0-9]+_labels.jpg shows—

validation batch labels validation batch labels

Train batch mosaics and labels

train_batch[0-9]+.jpg shows—

train batch mosaics and labels train batch mosaics and labels

Best Result

MBSTU_Underrated: 0.1346