search

rkotimi / cuddly-broccoli

none
0 stars 0 forks source link

CLOCs question #1

Open jiemingcui opened 3 years ago

jiemingcui commented 3 years ago

Hello! I have seen that you have achieved the CLOCs train on your own 3D & 2D detectors. Could you please have a talk with me about how to switch the code for us own proposals?

rkotimi commented 3 years ago

Hello @Cjmmmmmm ! I fuse the candidates from PointRCNN and Cascade RCNN and I can tell you the general steps. First, I replace final_box_preds and final_scores in function train_stage_2 of voxelnet.py with candidates and confidence scores from PointRCNN, respectively. Then I change the number 70400 to 100. After that, I train CLOCs and save the output of fusion_layer (cls_pred) as the rectified raw confidence scores of candidates from PointRCNN. Finally, I run the post processing code of PointRCNN to obtain the detection results.

jiemingcui commented 3 years ago

Thank you for your sharing. I want to train the CLOCs model with only the 3d proposals and 2d proposals, and I have met a lot of problems. Could you please share your code with me? I really really want to fix the problems now. Thank you!

------------------ 原始邮件 ------------------ 发件人: "rkotimi/cuddly-broccoli" @.>; 发送时间: 2021年8月3日(星期二) 下午5:51 @.>; @.**@.>; 主题: Re: [rkotimi/cuddly-broccoli] CLOCs question (#1)

Hello @Cjmmmmmm ! I fuse the candidates from PointRCNN and Cascade RCNN and I can tell you the general steps. First, I replace final_box_preds and final_scores in function train_stage_2 of voxelnet.py with candidates and confidence scores from PointRCNN, respectively. Then I change the number 70400 to 100. After that, I train CLOCs and save the output of fusion_layer (cls_pred) as the rectified raw confidence scores of candidates from PointRCNN. Finally, I run the post processing code of PointRCNN to obtain the detection results.

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub, or unsubscribe. Triage notifications on the go with GitHub Mobile for iOS or Android.