Xiujun Li

I used this codebase and pre-trained models, https://github.com/peteanderson80/bottom-up-attention, it is pre-trained on VG with 1600 tags.

Thanks for all. @zero-you I did a quick check the branch, that (VinVL updates) might be the cause to contaminate the master branch (for Oscar). Thanks for all.

@CavendishBosh @arensis-julia @zero-you I checked, you may can use the [git version](https://github.com/microsoft/Oscar/commit/9b07b6735267c0eb30fc2efd482d4295d13c7b4f) if you want to use Oscar (not VinVL), thanks.

Hi, First, 74.82 is on 2k val set, 76.12 is on (Server) test-std set (we do not have ground truth), which are not comparable ... Second, we release the [our...

Very soon, it is likely next Tuesday after the meeting.

Sorry for the late reply, not finalize it in the meeting of last Tuesday, trying to get one version around ECCV camera-ready (this Friday or next Tuesday's meeting).

@HimariO @RyanHuangNLP No update, it is currently used internally, probably will delay this for next new version. (Sorry for this, out of control.)

We will release a new version after the CVPR deadline.

See here, https://github.com/microsoft/Oscar/blob/master/VinVL_MODEL_ZOO.md#Oscarplus-pretraining