pyun-ram

> Yes, it is. I haven't done detailed research yet. For the same configuration, my model somehow cannot even work without timesteps. It seems like the timestamp introduces some bias...

I agree with youi @jinyu121

Still not yet...

The minimum overlap threshold for 2D evaluation on the image/ground plane and 3D evaluation is indeed 0.7, 0.5, 0.5 for car class. Please check again. :)

In MIN_OVERLAP Matrix: {{0.7, 0.5, 0.5}, {0.5, 0.25, 0.25}, {0.5, 0.25,0.25}}, each row is for a specific task: 2D evaluation on the image plane, 2D evaluation on the ground plane...

Indeed, I want to compute the KFAC for the parameters of Conv2d, Linear and BatchNorm2d layer. Is it possible to achieve this?

Thanks for the prompt reply! Yep, I have not seen some paper discussing the KFAC calculation for BN either... To get the KFAC for the supported layers (Linear and Conv2d),...

That's a great relief! :)

Thanks for your guidance! :) A pull request has been raised. A simple test case is also added to test the result and mode checking. I want to ask one...

A visualization example to compare this feature with GGN: ```python import torch from matplotlib import pyplot as plt from torch.nn import BatchNorm1d, CrossEntropyLoss, Flatten, Linear, Sequential from backpack import backpack,...