Logo classifier written utilizing OpenMMLab MMOCR, MMClassification tools and PyTorch.
Before installing the requirements, setup anaconda and create a new conda environment via
conda create -n logo_cls python=3.7 -y
conda activate logo_cls and install PyTorch via
conda install pytorch=1.8.0 torchvision=0.2.1 torchaudio=0.8.0 cudatoolkit=11.1 -c pytorchRun req.sh in the activated conda environment to setup requirements.
You can follow a procedure similar to this guide to install OpenMMLab libraries.
Download pretrained models for the logo classifier and place them into pretrains folder.
For logo classification, perform the steps in Augmentations & Split.ipynb and LogoClassifier.ipynb files consequently. Organize a logo_data directory where the subdirectories named according to companies. These subdirectories should include logo images of specified companies. Adjust the path for logo_data folder in Augmentations & Split.ipynb file.
For custom datasets, OpenMMLab new dataset guide is followed. The custom dataset file provided is datasets/logolist.py and it should be moved to mmcls/datasets folder and the dataset name should be added to__init__.py file once the required openmmlab libraries are installed.
For OCR tool, perform the steps in OCR Tool.ipynb file.