mmpretrain Roadmap of MMClassification

We keep this issue open to collect feature requests from users and hear your voice. Our monthly release plan is also available here.

You can either:

Suggest a new feature by leaving a comment.
Vote for a feature request with 👍 or be against with 👎. (Remember that developers are busy and cannot respond to all feature requests, so vote for your most favorable one!)
Tell us that you would like to help implement one of the features in the list or review the PRs. (This is the greatest things to hear about!)

Jul 13 '20 13:07 hellock

warm up and more pretrained models

Jul 16 '20 18:07 GoldenAbel

tricks like mix-up, auto-augmentation, and etc..

Jul 17 '20 07:07 LeeJZh

convert model to onnx.

Jul 17 '20 07:07 Tshzzz

EfficientNet, that could serve as backbone for mmdetection as well

Jul 20 '20 11:07 SilvioGiancola

some Nerual Architecture Search algorithms which could be implement on one GPU, like ENAS

Jul 21 '20 08:07 Ironteen

multi-brach for multi-task learning

Jul 22 '20 12:07 soldier828

convert model to onnx.

I think pytorch 1.5 supports it. torch.onnx

Jul 23 '20 02:07 LeeJZh

darknet networks and data aug, like darknet53, mosaic, cutmix, mixup

Jul 28 '20 09:07 GoldenAbel

benchmark on cifar10 and cifar100

Aug 04 '20 09:08 myopengit

Could you support HRNet?

Sep 20 '20 07:09 zhongqiu1245

Multi-label classification

Oct 10 '20 10:10 ousinkou

More types of loss supported, such as center loss, amsoftmax loss, and so on.

Dec 04 '20 07:12 Kongsea

focal loss pls！

Dec 06 '20 15:12 zhongqiu1245

vovnet-v2, it is faster and more accurate than resnet in my experience.

Dec 19 '20 10:12 hiyyg

HS-ResNet: Hierarchical-Split Block on Convolutional Neural Network https://arxiv.org/abs/2010.07621 There is a reference repo: https://github.com/bobo0810/HS-ResNet, but no pretrained models are provided. It would be nice if mmcls could include this and provide some pretrained models.

Dec 21 '20 11:12 hiyyg

The functionality of adding custom hooks like it is in mmdetection. I can take it up, if it seems like a good feature to the mmlab team.