Skip to content

Latest commit

 

History

History
50 lines (50 loc) · 2.17 KB

ssdd.md

File metadata and controls

50 lines (50 loc) · 2.17 KB
Raw

Using JDet with SSDD/SSDD+

Using JDet with SAR Ship Detection Dataset (SSDD/SSDD+).

Data Preparing

Downloading SSDD at https://github.com/TianwenZhang0825/Official-SSDD, and save to $SSDD_PATH$ as:

$SSDD_PATH$
└── Official-SSDD-OPEN
      ├──...
      ├──BBox_SSDD/voc_style
      |     ├──...
      |     ├──JPEGImages_train
      |     ├──JPEGImages_test
      |     ├──Annotations_train
      |     └──Annotations_test
      └──RBox_SSDD/voc_style
            ├──...
            ├──JPEGImages_train
            ├──JPEGImages_test
            ├──Annotations_train
            └──Annotations_test

Only the 8 folders above(JPEGImages_train, JPEGImages_test, Annotations_train, Annotations_test) are used in JDet.

Data Preprocessing

We need to rescale each image into a consistent size before training and testing.

cd $JDet_PATH$

We can set how the SSDD/SSDD+ is preprocessed by editing the configs/preprocess/ssdd_preprocess_config.py or configs/preprocess/ssdd_plus_preprocess_config.py, we use ssdd_plus_preprocess_config.py as an example:

type='SSDD+'
resize = 800
source_dataset_path='/home/cxjyxx_me/workspace/JAD/SAR/datasets/Official-SSDD-OPEN/RBox_SSDD/voc_style'
target_dataset_path=f'/home/cxjyxx_me/workspace/JAD/SAR/datasets/processed_SSDD_plus/'
convert_tasks=['test', 'train']

We need to set source_dataset_path to $SSDD_PATH$/Official-SSDD-OPEN/RBox_SSDD/voc_style, and set target_dataset_path to $PROCESSED_SSDD_PLUS_PATH$. Then we can set the resize paramters through resize. Finally, run the following script for preprocessing：

python tools/preprocess.py --config-file configs/preprocess/ssdd_plus_preprocess_config.py

For the way of configuring the processed SSDD/SSDD+ dataset in the model config file, please refer to $JDet_PATH$/projects/s2anet/configs/s2anet_r50_fpn_1x_ssdd.py/$JDet_PATH$/projects/s2anet/configs/s2anet_r50_fpn_1x_ssdd_plus.py:

dataset = dict(
    ...
)

Note: we rename the 'test' of SSDD/SSDD+ to 'val'.

Data Postprocessing

The Runner.val() in JDet will automatically calculate the AP50 of SSDD/SSDD+ test set.