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

Models and results in Model Zoo #38

Closed charmeleonz closed 1 year ago

charmeleonz commented 1 year ago

Dear author, in the model zoo, I am confused by descriptions in the model zoo. Specifically, for models in rows of "kitti train", did they not use eigen data? - which in turn means these models are trained in a fully supervised way without the CMKD framework?

Cc-Hy commented 1 year ago

Hi, the models in the model zoo are all trained with the CMKD framework. For the kitti train samples, we just treat them as unlabeled.

charmeleonz commented 1 year ago

That makes sense, thanks!

charmeleonz commented 1 year ago

@Cc-Hy Hi, I'm curious why the KITTI val results by the 'kitti train' (that is, without eigen clean data) are even inferior to CaDDN's original val results? This seems to mean distillation and whatever LiDAR-induced knowledge even hurt CaDDN's performance. Could you kindly share your insights into this?

xiaoxusanheyi commented 1 year ago

@Cc-Hy嗨,我很好奇为什么“kitti train”(即没有本征清洁数据)的 KITTI val 结果甚至不如 CaDDN 的原始 val 结果?这似乎意味着蒸馏和任何 LiDAR 诱导的知识甚至会损害 CaDDN 的性能。您能否分享您对此的见解?

你好,这个CMKD的model zoo 中kitti train”(即没有本征清洁数据)second那个数据是验证集上的3DAP吗?

charmeleonz commented 1 year ago

@Cc-Hy嗨,我很好奇为什么“kitti train”(即没有本征清洁数据)的 KITTI val 结果甚至不如 CaDDN 的原始 val 结果?这似乎意味着蒸馏和任何 LiDAR 诱导的知识甚至会损害 CaDDN 的性能。您能否分享您对此的见解?

你好,这个CMKD的model zoo 中kitti train”(即没有本征清洁数据)second那个数据是验证集上的3DAP吗?

Yes I believe so.

xiaoxusanheyi commented 1 year ago

@Cc-Hy嗨,我很好奇为什么“kitti train”(即没有本征清洁数据)的 KITTI val 结果甚至不如 CaDDN 的原始 val 结果?这似乎意味着蒸馏和任何 LiDAR 诱导的知识甚至会损害 CaDDN 的性能。您能否分享您对此的见解?

你好,这个CMKD的model zoo 中kitti train“(即没有本征清洁数据)second那个数据是验证集上的3DAP吗?

是的,我相信是的。

感谢!还有一个问题就是说CMKD 该如何可视化呢 就想原文图8的可视化图片那样,该使用那条指令呢?