README.md

##Label Assignment Matters: A Gaussian Assignment Strategy for Tiny Object Detection

RTSD dataset

Crop image dataset Download Google Drive
whole image dataset Baidu Drive (Passwd:nudt)

Trined model, logs and result file can be downloaded from the download link in the table.

Model	AP	AP_vt	Speed	Download
RetinaNet-S-GA	20.2	8.7	34.0	Google Drive Baidu Drive (Passwd:nudt)
FCOS-S-GA	19.6	7.9	34.8	Google Drive Baidu Drive (Passwd:nudt)
TTFNet-GA	21.8	10.3	34.3	Google Drive Baidu Drive (Passwd:nudt)
TTFNet-MiTB1-GA	24.2	10.4	37.6	Google Drive Baidu Drive (Passwd:nudt)

Install

Code (based on mmdetection) The detailed installation steps are in the \docs\get_started.md

Requirements

pytorch = 1.10.0
Linux or macOS (Windows is in experimental support)
Python 3.6+
PyTorch 1.3+
CUDA 9.2+ (If you build PyTorch from source, CUDA 9.0 is also compatible)
GCC 5+
numpy = 1.21.2
mmcv-full >=1.3.17 
mmdet = 2.19.0

You can also use this command

pip install -r requirements.txt

1. Install mmcv-full.

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/{cu_version}/{torch_version}/index.html

pip install mmcv-full -f https://download.openmmlab.com/mmcv/dist/cu110/torch1.7.0/index.html

2. Install MMDetection.

cd GuassionAssignment
pip install -r requirements/build.txt
pip install -v -e .  # or "python setup.py develop"

How to use?

1. Download the AI-TOD Dataset
2. Install mmdetection
3. Download our training models
4. Edit the data_root, in config files in ./configs_GA/

👇 Core File 👇

Config file

config_GA/atss_darknet53_aitod_2x_ga.py.
config_GA/fcos_darknet53_ga_aitod_2x.py. config_GA/retina_darknet53_aitod_2x_ga.py
config_GA/ttfnet_darknet53_aitod_iou_mask_ban_2x.py config_GA/ttfnet_mitb1_aitod_160k_ctfocal2.py

How to train?

python train.py ../config_GA/atss_darknet53_aitod_2x_ga.py 

How to test?

python test.py ../config_GA/atss_darknet53_aitod_2x_ga.py ../{your_checkpoint_path} --eval bbox