search

Cc-Hy / CMKD

Cross-Modality Knowledge Distillation Network for Monocular 3D Object Detection (ECCV 2022 Oral)
Apache License 2.0
107 stars 9 forks source link

About training strategy #54

Open ksh11023 opened 11 months ago

ksh11023 commented 11 months ago

Hello, I have a question regarding depth pre-training strategy.

step1. 60 epochs on KITTI train
step2. 20 epochs on KITTI train

Before running step1 and step2, do you only pre-train depth estimation part with eigen clean split for 10 epochs? for both CMKD and CMKD* results?

Thank you.

xiaoxusanheyi commented 11 months ago

您好， 我有一个关于深度预训练策略的问题。

步骤1。KITTI 训练 步骤 2 上有 60 个 epoch。KITTI 列车上的 20 个 epoch

在运行步骤 1 和步骤 2 之前，是否仅使用 eigen clean split 预训练 10 个 epoch 的深度估计部分？对于 CMKD 和 CMKD* 结果？

谢谢。 我也想知道呢

Cc-Hy commented 8 months ago

Hi, @ksh11023 Yes, when submitting the results to the test server, the backbones are depth pre-trained. When doing the ablation studies, this depth pre-training is not used.

TimGor1997 commented 6 months ago

Hi, @ksh11023 Yes, when submitting the results to the test server, the backbones are depth pre-trained. When doing the ablation studies, this depth pre-training is not used.

作者您好！请问是通过运行哪条指令可以实现depth pre-train呢？