TMP: Temporal Motion Perception with spatial auxiliary enhancement for moving Infrared dim-small target detection

Model

TMP

PR results on the DAUB and IRDST datasets

DAUB

IRDST

Datasets

1. Datasets are available at DUAB , IRDST and ITSDT

2. The COCO format need to convert to txt format.

python utils_coco/coco_to_txt.py

3. The folder structure should look like this:

Dataset
├─coco_train.txt
├─coco_val.txt
├─annotations
├─images
│   ├─train
│   │   ├─data5
│   │   │   ├─0.bmp
│   │   │   ├─ ...
│   │   │   ├─2999.bmp
│   │   │   ├─ ...
│   │   ├─ ...
│   ├─test
│   │   ├─data6
│   │   │   ├─0.bmp
│   │   │   ├─ ...
│   │   │   ├─398.bmp
│   │   │   ├─ ...
│   │   ├─ ...

3. The class of dataset should write to a txt file.

Such as model_data/classes.txt

Train

The hyper-parameters are set in train.py

python tarin.py

Evaluate

The hyper-parameters are set in vid_map_coco.py

python vid_map_coco.py

Visualization

The hyper-parameters are setted in vid_predcit.py

python vid_predcit.py

Reference

https://github.com/bubbliiiing/yolox-pytorch/