Fashion data to ReID format

Street2Shop

In order to transform raw Street2Shop dataset to COCO-ReID format follow the scripts in the directory as:

1. convert_to_jpg.py
2. street2shop2coco.py
3. street2shop_coco2reid.py

Convert to JPG

The script convert_to_jpg.py allows to easily convert all images in folder to JPG format.

NOTE If you want to transform raw Street2Shop dataset to ReID format, the conversion of images to JPG is necessary in order to use the subsequent scripts easily.

Example:

python convert_to_jpg.py \
--source-dir-path /path/to/s2s/images \
--target-dir-path /path/to/s2s/images_jpg \
--num-threads 8

Street2Shop to COCO

The result of the script is a reformatted train Street2Shop dataset in COCO format, but without ReID annotations.

Example:

python street2shop2coco.py \
--root-dir-path /path/to/s2s \
--metadata-dir meta \
--images-dir images_jpg \
--save-dir meta

Street2Shos to ReID

The result of the script is a reformatted full Street2Shop dataset in COCO format with ReID annotations. Images with bbox annotations will be cropped, resized to --target-image-size and saved to seperate folder. Images without bboxes will be only resized. Full split to val/test and query/gallery subsets will be produced allowing to train the CTLModel on it.

Example:

python street2shop_coco2reid.py \
--train-json-path /path/to/s2s/meta/all_street_train.json \
--root-dir-path /path/to/s2s \
--metadata-dir meta \
--images-dir images_jpg \
--save-dir meta_reid \
--target-image-size [320, 320] \
--minimum-bbox-area 1

Deep Fashion

Run deep_fashion2reid.py script. The dataset will be transformed to COCO-ReID format. Images with bbox annotations will be cropped, resized to --target-image-size and saved to seperate folder. Images without bboxes will be only resized.

NOTE To run the script without errors both low and high resolution images from the original dataset needs to be present in the directory.

Example:

python deep_fashion2reid.py \
--root-dir-path /path/to/s2s \
--target-image-size [320, 320]