search

shaunyuan22 / CFINet

The official implementation for ICCV'23 paper "Small Object Detection via Coarse-to-fine Proposal Generation and Imitation Learning"
Apache License 2.0
114 stars 5 forks source link

Extracting aug_hq_gt_bboxes #22

Closed 622tongtong closed 3 months ago

622tongtong commented 4 months ago

What's the feature?

The code about extracting features of aug_hq_gt_bboxes is as following: aug_hq_gt_rois = arb2roi([aug_hq_gt_bboxes], bbox_type=self.bbox_head.start_bbox_type) aug_hq_gt_roi_feats = self.aug_roi_extractor(x, aug_hq_gt_rois)

Through arb2roi, the image indexes of all aug_hq_gt_rois are 0, however, these gt bboxes are from different images. It may be my misunderstanding. Could you please provide an explanation?

Any other context?

No response

shaunyuan22 commented 3 months ago
  1. i guess you mean bbox2roi instead of arb2roi cause there is no arb2roi in our codes;
  2. if you look line 345 of feature_imitation_roi_head as follows: https://github.com/shaunyuan22/CFINet/blob/2167eebdc420046165e78a7555e40b044355cb58/mmdet/models/roi_heads/feature_imitation_roi_head.py#L345 you will find all the hq_gt_rois and hq_gt_roi_feats are processed at image level and we only handle one image at a time, hence the indexes are 0 and actually, the indexes here make no sense cause they are not use at all .