search

fanq15 / FewX

FewX is an open-source toolbox on top of Detectron2 for data-limited instance-level recognition tasks.
https://github.com/fanq15/FewX
MIT License
346 stars 48 forks source link

final_split_voc_10_shot_instances_train2017.json #5

Open Trepassing opened 4 years ago

Trepassing commented 4 years ago

Thanks for your code! I want to train on my datasets. However I am not aware of how final_split_voc_10_shot_instances_train2017.json is made. Could you share the code to me? Thanks a lot again!

fanq15 commented 4 years ago

The support quality is essential to few-shot learning. Therefore, I picked some supports with good qualities (e.g. proper box size, without occlusion and object diversity in the same class). Generally, I first select supports within a box size range and then pick the supports. I will upload the support selection file later for your reference.

Trepassing commented 4 years ago

thank you for your reply.

fanq15 commented 4 years ago

You can use 5_voc_part.py and 6_voc_few_shot.py to generate the final_split_voc_10_shot_instances_train2017.json. But it is different from the version in the github which is manually picked based on the generated annotations from the scripts (You can generate many annotations using the scripts and pick some good ones with the proper size and without occlusion).

Trepassing commented 4 years ago

Thanks again, much appreciated！I will have a try.