Visual Question Answering

• 이미지를 보고 주어진 질문에 답변하는 Visual Question Answering 모델 개발

Members

| 권태양 | 류재희 | 박종헌 | 신찬엽 | 조원 |

Set up

1. Install Requirements

$ pip install -r requirements.txt

2. Train

$ python3 train_v1.py # version 1
$ python3 train_v2.py # version 2
$ python3 train_v3.py # version 3

3. Inference

$ python3 inference_v1.py # version 1
$ python3 inference_v2.py # version 2
$ python3 inference_v3.py # version 3
$ python3 ensemble.py # ensemble

Code File

$ fashion_reader
├── config
│   ├── train_config_base.yaml
├── models
│   ├── get_model.py
│   └── vqa_model.py
├── modulus
│   ├── dataset.py
│   ├── earlystoppers.py
│   ├── recorders.py
│   ├── trainer.py
│   └── utils.py
├── results
├── train_v1.py
├── inference_v1.py
├── train_v2.py
├── inference_v2.py
├── train_v3.py
├── inference_v3.py
└── ensemble.py

Output

$ fashion_reader
└── results
    ├── train_v1
    │     ├── loss.png
    │     ├── model.pt
    │     ├── answers.csv
    │     ├── score.jpg
    │     ├── train_config_base.yaml
    │     └── train_log.log
    ├── train_v2
    │     └── ...
    └── train_v3
          └── ...

Description

Version	Pre-trained Model	Config
V1	xlm-roberta-base & resnet50	Link
V2	xlm-roberta-large & resnet50	Link
V3	xlm-roberta-base & resnet152	Link