DETR model for multi-mnist dataset


The directory structure of the code is as follws

.
├── datasets
│   ├── __init__.py
│   ├── mmnist
│   │   ├── test
│   │   │   └── normal
│   │   │       ├── bboxes.pickle
│   │   │       ├── filenames.pickle
│   │   │       ├── imgs
│   │   │       └── labels.pickle
│   │   └── train
│   │       └── normal
│   │           ├── bboxes.pickle
│   │           ├── filenames.pickle
│   │           ├── imgs
│   │           └── labels.pickle
│   └── mmnist.py
├── DETR_mmnist_inference.ipynb
├── DTER_mmnist_train.ipynb
├── models
│   ├── backbone.py
│   ├── detr.py
│   ├── __init__.py
│   ├── __init__.pyc
│   ├── matcher.py
│   ├── position_encoding.py
│   ├── transformer_attn.py
│   ├── transformer.py
│   └── util
│       ├── box_ops.py
│       ├── misc.py
│       └── __pycache__
│           ├── box_ops.cpython-36.pyc
│           └── misc.cpython-36.pyc
├── README
└── utils.py

Need pytorch 1.5 or later python2. Please refer requirements.txt for further detail.

The inference code is written in DETR_mmnist_inference.ipynb
The training code is written in DTER_mmnist_train.ipynb
The model contains the same code as given in official facebook research repository.
The datasets contain code for the dataloader.