search

NVlabs / ODISE

Official PyTorch implementation of ODISE: Open-Vocabulary Panoptic Segmentation with Text-to-Image Diffusion Models [CVPR 2023 Highlight]
https://arxiv.org/abs/2303.04803
Other
860 stars 49 forks source link

Evaluation of open-vocabulary instance segmentation on the Table B.5? #33

Open thangnn123456 opened 1 year ago

thangnn123456 commented 1 year ago

Dear @shalinidemello @xvjiarui,

Could you provide more details about how to reproduce the results in Table B.5? e.g. do we need to train on this instance task or just use the panoptic pretrained model? Could you provide the training and evaluation code? Is this from this line ?

Thank you!

xvjiarui commented 1 year ago

Hi @thangnguyennngoc ,

Since in Table B.5, we are evaluating recalls, we just use all the masks. The evaluation code follows the standard Recall calculation in COCO.