fastdup supports natively coco annotations. The example below shows how to run bounding box annotations in coco format. (Segmentation masks are not supported yet).
# let's get some small dataset to play with !git clone https://github.com/chongruo/tiny-coco.git import fastdup # create a fastdup object # input_dir is the location of the images referred in the annotation file. Absolute path please! # work_dir is the folder containing the resulting outputs fd = fastdup.create(work_dir='tiny-coco3', input_dir='/kaggle/working/tiny-coco/small_coco/train_2017_small/') fd.run(annotations='tiny-coco/small_coco/instances_train2017_small.json') #run any of fastdup galleries fd.vis.outliers_gallery()
Full example notebook is here: https://www.kaggle.com/code/graphlab/fastdup-coco-format-tutorial
Updated about 1 month ago