Type of annotations
-Depth Prediction Evaluation
-Object Detection (7481 training images and 7518 test images)
-Road/Lane Detection (289 training and 290 test images)
-Semantic and Instance Segmentation Evaluation (200 training images as well as 200 test images)
[x] Public dataset 조사를 통해 기존 데이터셋의 volume을 조사한다.
[x] 데이터셋 서베이 후, 우리데이터셋을 어떻게 보완할지 결정한다.
[ ] 기존 데이터셋의 volume에 준하는 우리만의 데이터 셋을 구축한다.
[x] 어떻게 라벨링 작업을 진행할지 계획하고, 그 계획에 적합하도록 기존 어노테이터를 변경한다.
Public dataset research
The KITTI Vision Benchmark Suite : http://www.cvlibs.net/datasets/kitti/ Python : https://github.com/utiasSTARS/pykitti
Cityscapes : https://www.cityscapes-dataset.com
현재 우리의 데이터 셋
문제점&보완방법
TFRecordsXML or json