brianjang / anything_color_recognition

This project implements the Vehicle Color Recognition using Convolutional Neural Network. I will train it with all types of data to recognize color of anything. Not sure whether it works or not!

Anything Color Recognition

In this repository, the codes for training anything color recognition are provided. This is a pytorch implementation of following paper: https://arxiv.org/pdf/1510.07391.pdf named Vehicle Color Recognition using Convolutional Neural Network

The colors are as follows: [Black, Blue, Brown, Green, Pink, Red, Silver, White, Yellow].

In this repo I try to train a model to recognize the color of anything. Right now the model is trained on vehicle dataset and fabric dataset. I will be adding more objects to the dataset.

Dataset

The dataset is mainly based on vehicle dataset. The rest are also being added to the dataset.

Installation:

pip install -r requirements.txt

Hyper parameters

The hyper-parameters are set in the settings.py module. You can adjust them based on your need.

Train

python train.py

Graphs

cd output/exp_4/finetune/lightning/version_0/
tensorboard --logdir .

inference.py

python inference.py --model_path output/exp_1/best.ckpt --labels_map_path output/exp_1/labels_map.pkl --img_path sample_images/sample_01.jpg

References:

1. https://arxiv.org/pdf/1510.07391.pdf
2. https://github.com/pooya-mohammadi/deep_utils