search

hanoonaR / object-centric-ovd

[NeurIPS 2022] Official repository of paper titled "Bridging the Gap between Object and Image-level Representations for Open-Vocabulary Detection".
Apache License 2.0
284 stars 19 forks source link

Reproduce the cross-dataset evaluation #20

Closed Changes92 closed 1 year ago

Changes92 commented 1 year ago

Thank you for your great work!

Recently, I hoped to reproduce the cross-data evaluation but failed to obtain the reported results. Could you please provide the following files? 1) the YAML config file, 2) the files in ./ovd/datasets/, and 3) the CLIP word embedding?

Thank you so much!

hanoonaR commented 1 year ago

Hi @tmp12316 ,

Thank you for showing your interest in our work. Please note the performance we report on OpenImages and Objects365 cross-dataset is the zero-shot evaluation of our best LVIS model, which is the Detic-based CenterNet model. We have not included these models in our repository and therefore sharing the requested files wouldn't be helpful. But please let us know if you would like to know any implementational details.

Thank you.