Furniture-Classification

iMaterialist Challenge (Furniture) at FGVC5, Kaggle Competition

Data Downloader

We have download.py for training data, downloadd_valid.py for validation data and download_test.py for testing data, an example running is showed below.

python download.py

Data Preprocess

1. For pytorch model, use dataset_gen.py for making 128 directories for training set and validation set and putting the valid images into these folders according to their labels. preprocess.py is used to putting all valid image files into the newly created folder.
2. For tensorflow model, run preprocess_main.py to preprocess the images and store them in npy file. Also, download AlexNet weights, and save the bvlc-alexnet.npy

Training

1. For pytorch model

python transfer_main.py

The weights are stored in best_weight.pth. 2. For tensorflow model, open pipeline.ipynb for training.

Testing

python predict.py

to load the weight and make prediction on testing set, the result of prediction will be stored in test_prediction.pth.

Submission File for Kaggle

python gen_csv.py

to map index of training with testing and generating csv file for submission, in submission.csv.