Skip to content

Latest commit

 

History

History
79 lines (63 loc) · 2.62 KB

README.md

File metadata and controls

79 lines (63 loc) · 2.62 KB
Raw

Prepare Datasets for DVIS++

A dataset can be used by accessing DatasetCatalog for its data, or MetadataCatalog for its metadata (class names, etc). This document explains how to setup the builtin datasets so they can be used by the above APIs. Use Custom Datasets gives a deeper dive on how to use DatasetCatalog and MetadataCatalog, and how to add new datasets to them.

DVIS++ has builtin support for a few datasets. The datasets are assumed to exist in a directory specified by the environment variable DETECTRON2_DATASETS. Under this directory, detectron2 will look for datasets in the structure described below, if needed.

$DETECTRON2_DATASETS/
  ytvis_2019/
  ytvis_2021/
  ovis/
  VIPSeg/
  VSPW_480p/

You can set the location for builtin datasets by export DETECTRON2_DATASETS=/path/to/datasets. If left unset, the default is ./datasets relative to your current working directory.

The model zoo contains configs and models that use these builtin datasets.

Expected dataset structure for YouTubeVIS 2019:

ytvis_2019/
  {train,valid,test}.json
  {train,valid,test}/
    Annotations/
    JPEGImages/

Expected dataset structure for YouTubeVIS 2021:

ytvis_2021/
  {train,valid,test}.json
  {train,valid,test}/
    Annotations/
    JPEGImages/

Expected dataset structure for Occluded VIS:

ovis/
  annotations/
    annotations_{train,valid,test}.json
  {train,valid,test}/

Expected dataset structure for VIPSeg:

After downloading the VIPSeg dataset, it still needs to be processed according to the official script. To save time, you can directly download the processed VIPSeg dataset from baiduyun (password is dvis).

VIPSeg/
  VIPSeg_720P/
    images/
    panomasksRGB/
    panoptic_gt_VIPSeg_{train,val,test}.json

Expected dataset structure for VSPW:

VSPW_480p/
  data/
  {train,val,test}.txt

Register your own dataset:

  • If it is a VIS/VPS/VSS dataset, convert it to YTVIS/VIPSeg/VSPW format. If it is a image instance dataset, convert it to COCO format.
  • Register it in /dvis_Plus/data_video/datasets/{builtin,vps,vss}.py