Siddhartha

Any plans to add mixed precision training ?

1

Pytorch 1.6 now natively supports [mixed precision training](https://pytorch.org/docs/stable/notes/amp_examples.html). This can lead to much faster training without loss in model accuracy. I was wondering if you have any plans to add...

This can be nice issue for beginners. For reference, existing Thread safe implementations are [here](https://github.com/ZigRazor/CXXGraph/blob/master/include/Graph/Graph_TS.hpp#L343). Similar pattern can be used for adding thread Safe implementations of Boruvka, Prim & Kruskal...

Hi. In the paper you derived equations for the case when softmax function is used to obtain final class probabilities. (Equations 21-24) I am unable to find an implementation of...

Effect of DropBlock and RFB block

1

Hi. Thanks for sharing the code. I had two questions. In section 4.4 of the paper you mentioned that , > final model is YOLOv3 with ASFF*, which is an...

fix bug due to linspace float argument

3

When I tried to evaluate your model checkpoint on kitti dataset, I got the following error: `TypeError: 'numpy.float64' object cannot be interpreted as an integer` This issue also cropped up...

Hi I was going through the region_loss code. I found a possible bug in build_targets function, ```python for t in range(nTrueBox): if target[b][t][1] == 0: break ``` I think it...

STN shouldn't be used for semantic segmentation

3

Hi. I see that in this repo, STN is used for [ semantic segmentation case also](https://github.com/fxia22/pointnet.pytorch/blob/f0c2430b0b1529e3f76fb5d6cd6ca14be763d975/pointnet/model.py#L104). According to the author, for semantic segmentation T-Net mayn't be able to find a...