AI-Convex-Solutions / car-recognition

Resnet152 Car Recognition Model. Returns manufacturer, year manufactured, model, color.

Home Page:https://cardetect.tech/

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

DEMO

www.cardetect.tech

Car Recognition (English)

Model

The model is trained using 300,000+ car photos. The model returns:

  • Car manufcaturer (audi, bmw, etj.) - Accuracy 91%,
  • Manufactured year (2013, 2019) - Accuracy 33%,
  • Car model (Golf 5, Clio) - Accuracy 81%,
  • Car color (Red, Black) - Pretrained.

Requirements

  1. Download the file https://www.dropbox.com/s/95tefoi8uz4fhor/alpha_model?dl=0 and place it under: model/finished_models/alpha/

  2. Download the file https://www.dropbox.com/s/hbftvpih9g4rshc/final_model_85.pt?dl=0 place it under: model/colors/.

  3. Download the file https://www.dropbox.com/s/4orqou90l7fp97v/resnet152-f82ba261.pth?dl=0 and place it under the root repository ./resnet152-f82ba261.pth.

Usage

Start by building the dockerfile: docker build --tag car-recognition .. Then start docker: docker run -d -p 5000:5000 car-recogntion.

Requests

curl --location --request POST 'http://localhost:5000/predict' \
--form 'file=@"/home/kryekuzhinieri/Desktop/audi_q3_2016.png"'

Plans

  • [] Year should change from one value to multiple years.
  • Model should return color.
  • [] Model must have an accuracy of 90% in all categories.
  • [] User should be able to request multiple images at once.
  • [] Accept pixels and not images for security.

Car Recognition (shqip)

Modeli

Modeli është trajnuar duke përdorur fotografi të më shumë se 300.000 makinave. Çdo fotoje modeli i përgjigjet me:

  • Kompaninë e automjetit (audi, bmw, etj.) - Saktësia 91%,
  • Vitin e prodhimit (2013, 2019) - Saktësisa 33%,
  • Modelin e automjetit (Golf 5, Clio) - Saktësia 81%,
  • Ngjyrën e automjetit (Kuqe, Zezë) - Model i gatshëm.

Kushtet paraprake

  1. Shkarko skedarin e modelit nga https://www.dropbox.com/s/95tefoi8uz4fhor/alpha_model?dl=0 dhe vendose në: model/finished_models/alpha/

  2. Shkarko edhe skedarin tjetër https://www.dropbox.com/s/hbftvpih9g4rshc/final_model_85.pt?dl=0 dhe vendose në: model/colors/.

  3. Shkarko skedarin https://www.dropbox.com/s/4orqou90l7fp97v/resnet152-f82ba261.pth?dl=0 dhe vendose në dosjen bazë ./resnet152-f82ba261.pth.

Përdorimi

Fillimisht ndërtoje Dokerin: docker build --tag car-recognition .. Pastaj filloje dokerin: docker run -d -p 5000:5000 car-recogntion.

Kërkesat

curl --location --request POST 'http://localhost:5000/predict' \
--form 'file=@"/home/kryekuzhinieri/Desktop/audi_q3_2016.png"'

Planet

  • [] Viti duhet të ndryshohet nga 20132013, 2014, 2015 ose 2013-2015.
  • Modelit duhet t'i shtohet ngjyra.
  • [] Modeli duhet të ketë saktësi së paku 90% në të gjitha kategoritë.
  • [] API-ji duhet të lejojë disa imazhe në një kërkesë.
  • [] Pranimi i pikselave e jo i imazheve.

About

Resnet152 Car Recognition Model. Returns manufacturer, year manufactured, model, color.

https://cardetect.tech/

License:MIT License


Languages

Language:Python 97.4%Language:Dockerfile 2.6%