search

zhang-tao-whu / e2ec

E2EC: An End-to-End Contour-based Method for High-Quality High-Speed Instance Segmentation
Other
214 stars 45 forks source link

About detector #25

Closed hannn0919 closed 1 year ago

hannn0919 commented 1 year ago

Hi, thanks for your excellent work!

I tried to train this work on COCO dataset, but it converges very slowly and needs to train more than 100 epochs. I notice that you mention it is because of CenterNet(detector) in previous issue, and you rebuild this network in FCOS version.

I wonder when will the FCOS version be released? Or can you explain how do I change the detector? Which files and what functions should I change? Thanks! :)

zhang-tao-whu commented 1 year ago

Hi! I have implemented the fcos-based e2ec using mmdetection, and incorporates some minor improvements. Training 12 epochs without data augmentation on coco can achieve 32.9 AP on coco-val, and using data augmentation and training 48 epochs can achieve 35.1 AP on coco-val. You can [e2ec_mmdet]visit e2ec_mmdet. The config file is e2ec_fcos_center-normbbox-centeronreg-giou_r50_caffe_fpn_gn-head_dcn_feature64_FPNinit_p4evolve_bounddml_wholeInstance_reuniform_mspoints_attentive_normed-rela-coords_1x_coco.py and e2ec_fcos_center-normbbox-centeronreg-giou_r50_caffe_fpn_gn-head_dcn_feature64_FPNinit_p4evolve_bounddml_wholeInstance_reuniform_mspoints_attentive_normed-rela-coords_mstrain_4x_coco.py. You need to compile some components under csrc before you can use them. I don't have time to organize the code, so it may be difficult for you to read.

hannn0919 commented 1 year ago

Really appreciate your reply and patience!!! I'll try to handle this. Thanks again. :)