Skip to content

Latest commit

 

History

History
74 lines (52 loc) · 2.87 KB

DATA.md

File metadata and controls

74 lines (52 loc) · 2.87 KB
Raw

We provide guidance for preparing the data used by MQ-DET. Note that not all data are needed for a specific experiments. Please check the Required Data fields in README to download necessary data. All data should by placed under the DATASET folder.

The data should be organized in the following format:

DATASET/
    coco/
        annotations/
            lvis_od_train.json
            lvis_od_val.json
            lvis_v1_minival_inserted_image_name.json
        train2017/
        val2017/
        test2017/
    Objects365/
        images/
        zhiyuan_objv2_train.json
    odinw/
       AerialMaritimeDrone/
       ...
       WildfireSmoke/

Objects365

We found that the Objects365 v1 is unavailable now. Please try to download v2 as follows.

Download the Objects365 dataset from YOLOv5.

You can also use custom datasets for modulated pre-training as long as they are in COCO format.

LVIS

LVIS use the same images as COCO. Thus prepare the COCO images and annoations first and place them at DATASET/coco/.

All processed LVIS annotation files can be downloaded through:

train minival val 1.0
link link link

And place them at DATASET/coco/annotations/.

If you want to process by yourself rather than using the pre-processed files, please follow the instruction in GLIP, summarized as following.

Download the following annotation files:

    wget https://penzhanwu2bbs.blob.core.windows.net/data/GLIPv1_Open/coco/annotations/lvis_v1_minival_inserted_image_name.json -O DATASET/coco/annotations/lvis_v1_minival_inserted_image_name.json

    wget https://penzhanwu2bbs.blob.core.windows.net/data/GLIPv1_Open/coco/annotations/lvis_od_val.json -O coco/annotations/lvis_od_val.json"

Also download the training set for extracting vision queries:

   wget https://s3-us-west-2.amazonaws.com/dl.fbaipublicfiles.com/LVIS/lvis_v1_train.json.zip -O coco/annotations/lvis_v1_train.json.zip

Unpack the .zip file to coco/annotations/lvis_v1_train.json, and convert it to coco format:


python utils/add_file_name.py

Object Detection in the Wild (ODinW)

Download ODinW

python odinw/download_datasets.py

configs/odinw_35 contain all the meta information of the datasets. configs/odinw_13 are the datasets used by GLIP. Each dataset follows the coco detection format.

Please refer to GLIP for more details.