-
Hello, I read your paper and have several questions.
My question is about Section 4.2 / Appendix D - parallelized training.
I think I haven't fully understood why the use of ID Decoder enables par…
-
认真读了作者的论文,收获很多,感谢您的工作!
关于论文中的一些点的探讨:
1. Transform的结构使得不定长轨迹和检测预测的多输入处理变得可行,这也是设计“ID decoder”的重要支柱。从训练的角度看,因为拼接了ID-word的embedding,ID预测也就具有了直接的基础。这个任务使用监督学习和对比学习都比较适用,文中贴出的消融实验结果差异还挺大,感觉也可挖一下这个点
2. 在…
-
I follow the [installation Doc](https://github.com/open-mmlab/mmdetection/blob/master/docs/en/get_started.md), and failed when download config and checkpoint file:
```
mim download mmdet --config …
-
### Prerequisite
- [X] I have searched [Issues](https://github.com/open-mmlab/mmdetection3d/issues) and [Discussions](https://github.com/open-mmlab/mmdetection3d/discussions) but cannot get the exp…
-
First of all thank you for your team's work!
My own dataset has 3k training samples and I wish to train from scratch using res50, this facilitates subsequent changes to the different backbone netwo…
-
when I use the shell train_RLIP_ParSeDA_v2_mixed_vgcocoo365_swinL.sh to reproduce the pretraining process, I set the --pretrained to the OD weight I downloaded swin_large_cocoo365_bs64_lr141_drop_path…
-
Taking DN-DETR as example
# 1 add a dn_detr_onnx.py, you only need to change the `forward` function.
```
# coding=utf-8
# Copyright 2022 The IDEA Authors. All rights reserved.
#
# Licensed under…
-
Test: [4990/5000] eta: 0:00:01 class_error: 90.00 loss: 18.1650 (18.6082) loss_bbox_dn: 0.0000 (0.0000) loss_giou_dn: 0.0000 (0.0000) loss_ce_dn: 0.0000 (0.0000) loss_ce: 0.9389 (0.9540) loss…
1106X updated
5 months ago
-
Hi, thanks for bringing new insights to the DETR series. DN-DETR is really an excellent work that can get such high performance with only 12 epochs.
After reading the paper, I have several question…
-
您好,我想请问一下如何通过你这个模型将检测到的3D关键点坐标保存下来呢.