Mu Hu

> Hi Assran, thank you for your greatTTTTTTTTT work! I wonder if using a longer pre-training schedule (800/1200/1600 epochs), how much performance superiority can we get upon previous methods like...

I successfully re-implemented the performance on NYU but failed on Scannet. I have not yet found out the reasons.

> 你好： 我看文章里OREPA和RepVGG结合时，是直接在conv_3_3上加OREPA， 而不是直接将conv_3_3/conv_1_1/identity三个分支换成OREPA的形式。请问这样做的是因为直接将conv_3_3/conv_1*1/identity三个分支换成OREPA进行训练效果不好么？ 是的，这样并不是很优雅

> Hello, I'm very confused about accuracy of ResNet34. Specifically, I train ResNet34 many time, but accuracy of ResNet34 is about 74.40. I found that this paper and RepVGG both...

> Hi, thanks for your great work! > > I tried to reproduce the visualization of branch-level similarity of OREPA blocks, but the unexpected results emerged. > > Could you...

We have posted it in the supplementary materials. (On an a100 gpu.) 

> I wondered what camera-related parameters matter most when using the `vit` models and if there are any guidelines to set those parameters. What do you mean by camera-related parameters?...

You need to change nothing but the focal length, which should align with the real case.

> I changed the intrinsic accordingly to match it to my custom image: https://github.com/YvanYin/Metric3D/blob/main/mono/utils/do_test.py#L258. Is this the only change in the code to get the metric depth on a custom...

一周左右？这个模型看上去很小，但在a100上跑起来跟large速度差不多，而且只在室外数据集上训过，泛化性一般