search

akshitac8 / OW-DETR

[CVPR 2022] Official Pytorch code for OW-DETR: Open-world Detection Transformer
235 stars 39 forks source link

store exemplars #5

Closed aooating closed 2 years ago

aooating commented 2 years ago

Where in the code do we store 50 samples per known category?

akshitac8 commented 2 years ago

Hello @aooating We don't store 50 samples per category for OW-DETR.

aooating commented 2 years ago

In C. Additional Implementation Details, there is writing in the incremental step the learning rate is reduced by a factor of 10 and trained using a set of 50 stored exemplars per known class.

akshitac8 commented 2 years ago

The 50 stored exemplars are the images used for fine-tuning which are present in _ft.txt files.

aooating commented 2 years ago

Thank you for your reply! I would also like to know how to generate those ft.txt files. Looking forward to your reply!

akshitac8 commented 2 years ago

@aooating the files are already present inside these paths https://github.com/akshitac8/OW-DETR/tree/main/data/OWDETR/VOC2007/ImageSets/ https://github.com/akshitac8/OW-DETR/tree/main/data/OWOD/VOC2007/ImageSets