search

sabarim / STEm-Seg

This repository contains the official implementation of the paper "STEm-Seg: Spatio-temporal Embeddings for Instance Segmentation in Videos"
153 stars 23 forks source link

make dataset #14

Closed luoww1992 closed 3 years ago

luoww1992 commented 3 years ago

@Ali2500 i will use YouTube-VIS dataset to do it. i have make a new dataset like YouTube-VIS and i make the json file like its json file. i have some questions when i make the last standard json file like youtube_vis_train.json you provide。 1, i don't know how to make the value: segmentations in youtube_vis_train.json from annotation['segmentations'] ? 2,are the keys: bboxes, areas useful?

Ali2500 commented 3 years ago
  1. The segmentation values are created using the mask to RLE encoding function in pycocotools (see pycocotools.masktools.encode)
  2. I don't think the bboxes and areas are actually used.