search

GuangxingHan / QA-FewDet

Code for ICCV 2021 paper: 'Query Adaptive Few-Shot Object Detection with Heterogeneous Graph Convolutional Networks'
31 stars 6 forks source link

question about the Heterogeneous gcn #7

Open yomik-js opened 1 year ago

yomik-js commented 1 year ago

dear author, when I read the code, I found that when you construct the inter-class subgraph, you only update by calculating the similarity of each class prototype, without using the described GCN, and I just found an update for the base class category node, do not find the novel class node. Maybe it is my fault for not finding it, but can you help me? thank you very much!!!

GuangxingHan commented 1 year ago

Here is the code to invoke the inter-class subgraph. The function ss_edge() has two arguments support_feature_ls and support_cls_list, which are the features and class_ids of the few-shot classes sampled in the current episode. The goal of ss_edge() is to update the support features of these classes using the base classes memory stored in self.support_dict_base.

During meta-training, we only sample few-shot classes from base classes. But during few-shot fine-tuning, we sample few-shot classes from both base and novel classes. In this stage, we would update the novel classes support features using ss_edge().