search

alirezazareian / ovr-cnn

A new framework for open-vocabulary object detection, based on maskrcnn-benchmark
MIT License
229 stars 28 forks source link

Pretrained Models in the first stage #4

Closed lxtGH closed 3 years ago

lxtGH commented 3 years ago

❓ Questions and Help

Hi！Again, thanks for the code.

Could you release the pretrained models in the first stage for reference?

I follow the two stage pipeline. First train the model using COCO caption then training with coco zero setting.

However, I can not get the right results of the paper.

There is a huge drop on seen class mAP.

@alirezazareian

alirezazareian commented 3 years ago

Here is the pretrained model checkpoint. Please let me know if this doesn't resolve the issue.

lxtGH commented 3 years ago

Hi! Thanks for your reply. The gap has been resolved.