pytorch-Tiny-Darknet

PyTorch implementation of the Tiny Darknet Image Classification algorithm

This repository contains code for a classifier based on Tiny Darknet, implementedin PyTorch. The code is based on the official code of YOLO v3, as well as a PyTorch port of the original code, by ayooshkathuria. The aim of this project is to have pytorch compatible Tiny Darknet classifier.

As of now, the code only contains the Classification module.

Requirements

1. Python 3.5
2. OpenCV
3. PyTorch 0.4

Running the detector

On single or multiple images

python detect.py --image imgs/Malamute.jpg 

Output would be expected to be something similar like

Class = malamute  Confidence = 62.50366973876953
Class = Norwegian elkhound  Confidence = 17.33596420288086
Class = Eskimo dog  Confidence = 9.506247520446777
Class = Siberian husky  Confidence = 5.180300235748291
Class = keeshond  Confidence = 4.281615257263184

Inference Time Benchmarks

The average inference time observed on my machine (GTX 1050) is ~ 3ms