Dataset Viewer
View in Dataset Viewer
Viewer
The viewer is disabled because this dataset repo requires arbitrary Python code execution. Please consider
removing the
loading script
and relying on
automated data support
(you can use
convert_to_parquet
from the datasets library). If this is not possible, please
open a discussion
for direct help.
This repo provides a script to load refcoco, refcocog, and refcoco+ with local files. You should download the data first.
[DATA_ROOT]
|_ images
|_ train2014
|_ ... (images)
|_ refcoco
|_ instances.json
|_ refs(google).p
|_ refs(unc).p
|_ refcocog
|_ instances.json
|_ refs(google).p
|_ refs(umd).p
|_ refcoco+
|_ instances.json
|_ refs(unc).p
Usage:
REPO_NAME = "yiqun/referit"
dataset = "refcoco"
DATA_ROOT = ...
split = "train" # choices: train, val, test, testA, testB
split_by = "unc"
datasets.load_dataset(REPO_NAME, dataset, data_dir=DATA_ROOT, split=split, split_by=split_by)
- Downloads last month
- 14