search

shuangw98 / MFDC

Multi-Faceted Distillation of Base-Novel Commonality for Few-shot Object Detection, ECCV 2022
MIT License
27 stars 4 forks source link

Question about queue update #6

Closed vineetrshenoy closed 1 year ago

vineetrshenoy commented 1 year ago

I am running the code during novel fine-tuning on VOC split 1, 1-shot seed 0. During the memory bank update, it seems like the memory bank is updating more samples than there are in the novel set. See below at the debugging output

mfdc_question

As you can see, the uniq_c indices are [0, 9, 10, 17, 18]. The novel indices correspond to [15, 16, 17, 18, 19]. For index 17, features_c_s is of shape [26, 2048]. This seems to violate the 1-shot scenario.

I may be mis-understanding something. What is happening here?

shuangw98 commented 1 year ago

Hi, features_c_s indeed denotes the number of proposal features for each class. Any region proposals with IoU > threshold for an object cound be an element of features_c_s, so it's normal for the length of features_c_s to be greater than K under K-shot setting.

vineetrshenoy commented 1 year ago

Sorry for the late reply -- thanks, this makes sense. I am closing the issue.