Visual_Recognition_Deep_Learning_final_project

Final project "iNaturalist 2019 at FGVC6 Fine-grained classification spanning a thousand species" https://www.kaggle.com/c/inaturalist-2019-fgvc6

BBN: Bilateral-Branch Network with Cumulative Learning for Long-Tailed Visual Recognition with Shift-Invariant Convolutional Neural Network

This repository is based on the official PyTorch implementation of paper BBN: Bilateral-Branch Network with Cumulative Learning for Long-Tailed Visual Recognition.

Main requirements

• PyTorch ≥1.0
• torchvision ≥0.2.2_post3
• TensorboardX
• Python 3

Pretrain models for iNaturalist

We provide the BBN pretrain models for iNaturalist 2018: Google Drive

Usage

# To train long-tailed iNaturalist2019 with imbalanced ratio of 50:
python main/train.py  --cfg configs/iNaturalist2019.yaml     

# To validate with the best model:
python main/valid.py  --cfg configs/iNaturalist2019.yaml

You can change the experimental setting by simply modifying the parameter in the yaml file.

Data format

The annotation of a dataset is a dict consisting of two field: annotations and num_classes. The field annotations is a list of dict with image_id, fpath, im_height, im_width and category_id.

Here is an example.

{
    'annotations': [
                    {
                        'image_id': 1,
                        'fpath': '/home/BBN/iNat19/images/train_val2019/Plantae/7477/3b60c9486db1d2ee875f11a669fbde4a.jpg',
                        'im_height': 600,
                        'im_width': 800,
                        'category_id': 7477
                    },
                    ...
                   ]
    'num_classes': 8142
}

You can use the following code to convert from the original format of iNaturalist. The images and annotations can be downloaded at iNaturalist 2019

# Convert from the original format of iNaturalist
python tools/convert_from_iNat.py --file train2019.json --root iNat19/images --sp jsons