# fast-reid-naic2021-demo

**Repository Path**: pengpeng-yu/fast-reid-naic2021-demo

## Basic Information

- **Project Name**: fast-reid-naic2021-demo
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Apache-2.0
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 10
- **Forks**: 8
- **Created**: 2021-12-21
- **Last Updated**: 2022-04-08

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

### Fast Reid NAIC2021 Demo

A simple demo of NAIC2021 visual feature encoding preliminary first stage based on [fast-reid](https://github.com/JDAI-CV/fast-reid).
Step by Step training and inference procedures as follows.



#### 1. Set up with Conda

```shell script
conda create -n fastreid python=3.7
conda activate fastreid
conda install pytorch torchvision torchaudio cudatoolkit=10.2 -c pytorch
pip install -r docs/requirements.txt
```

#### 2. Download preliminary dataset

Download train.zip, test_A.zip from [official link](https://www.heywhale.com/home/competition/61b81042902a13001708eb17/content/2) and unzip them. Folders you will get:

```
train/
    train_feature/
    train_list.txt
test_A/
    gallery_feature_A/
    query_feature_A/
```

#### 3. Put the dataset in right place

Datasets should be placed in "datasets" folder of this project. You could move or link the folders.

Expected structure:
```bash
path/to/this/project/
    datasets/
        NAIC2021Reid/
            train_feature/
            gallery_feature_A/
            query_feature_A/
            sub_train_list.txt
            val_gallery_list.txt
            val_query_list.txt
```
File lists above are generated from official train_list.txt. They are necessary for training and validation and have been already placed in desired path.

File lists are not needed during inference.

#### 4. Training:

```shell
 python tools/train_net.py --config-file ./configs/NAIC2021Reid/sbs_mlp2x.yml MODEL.DEVICE "cuda:0"
 # Outputs at logs/NAIC2021Reid/sbs_mlp2x
```

#### 5. Inference 

```shell
python tools/train_net.py --config-file ./configs/NAIC2021Reid/sbs_mlp2x_inference.yml --infer-only MODEL.DEVICE "cuda:0"
# Submit file at logs/NAIC2021Reid/sbs_mlp2x_inference/Inference_On_NAIC2021ReidTestA/results.json
```