Jimmy Wang

problems about preprocessing on emotionet

1

1. about image: how to crop and resize image? I tried to use openface's alignment, the training result is good, but when apply on wild image, the model cannot tolerant...

Can you share the pretrained model on Kitti please? Thank you!

Corner case of box_nms

1

https://github.com/kuangliu/torchcv/blob/6291f3e1e4bbf6467fd6b1e79001d34a59481bb6/torchcv/utils/box.py#L128 When `overlap` contains only 1 element, `ids` will be a 0-dimensional tensor, which will cause dimension mismatch later. Examples: ``` >>> (torch.Tensor([0.1, 0.2, 0.6]) < 0.45).nonzero().squeeze() tensor([0, 1]) >>>...

Why use nonempty bins rather than all bins?

1

Why you divide `weights` by nonempty bins (`n`) rather than all bins(`self.bins`)?https://github.com/libuyu/GHM_Detection/blob/3647287710416c91077805d504349fb947c2e9bd/mmdetection/mmdet/core/loss/ghm_loss.py#L54 I think `M` is the amount of all bins in the paper. Am I missing something?

qm9_nn_conv.py gives unexpected large MAE on target 7

3

## ❓ Questions & Help When I run the example code [qm9_nn_conv.py](https://github.com/rusty1s/pytorch_geometric/blob/master/examples/qm9_nn_conv.py) , I found that the MAE of target 7 is much larger than it should be. The expected...