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

SODA-A config #10

Closed 622tongtong closed 7 months ago

622tongtong commented 7 months ago

Branch

master branch https://mmdetection.readthedocs.io/en/latest/

📚 The doc issue

I wonder if it is possible to upload the config file of obb task based on SODA-A. Cus there is only hbb config. Thank you so much for sharing.

Suggest a potential alternative/fix

No response

shaunyuan22 commented 7 months ago

Sry, this is not in our plan. The code structure of mmrot is highly similar to that of mmdet, and RotatedRPN outputs hbb proposals and the operation about region features in feature_imitation_roi_head.py has nothing to do with hbb or obb, which means just transfering the codes in crpn_head.py and feature_imitation_roi_head.py to rotated_rpn.py and rotate_standard_roi_head.py will work.

Alexwong-1024 commented 7 months ago

According to what you mean, I only need to modify your improved codes in crpn_head.py and feature_imitation_roi_head.py in mmdet, and modify them in the corresponding places of rotated_rpn.py and rotate_standard_roi_head.py in mmrotate? Then do I use the mmrotate module when training? Based on this modification, I can copy your configs/cfinet/faster_rcnn_r50_fpn_cfinet_1x.py file to SODA-mmrotate /configs/, and then use SODA-mmrotate’s train.py for training, or should you stay in the original CFINet file and use train.py for training? Looking forward to your reply. @shaunyuan22

622tongtong commented 7 months ago

OKay, thanks for your replying.

Alexwong-1024 commented 7 months ago

This is just my idea. I have just started in this direction. If you try it successfully, can you share how to modify it? Thank you！！！

shaunyuan22 commented 7 months ago

According to what you mean, I only need to modify your improved codes in crpn_head.py and feature_imitation_roi_head.py in mmdet, and modify them in the corresponding places of rotated_rpn.py and rotate_standard_roi_head.py in mmrotate? Then do I use the mmrotate module when training? Based on this modification, I can copy your configs/cfinet/faster_rcnn_r50_fpn_cfinet_1x.py file to SODA-mmrotate /configs/, and then use SODA-mmrotate’s train.py for training, or should you stay in the original CFINet file and use train.py for training? Looking forward to your reply. @shaunyuan22

exactly right, maybe it takes a few modifications when copying the config