takuoko
takuoko
**Describe the feature** I would like to be able to specify the config under .mim as the base config and run it as shown below. ``` _base_ = [ 'site-packages/mmcls/.mim/config/_base_/models/resnet18_cifar.py',...
## Motivation It is complicated to calculate the output channels of the backbone and set them as the input channels of the head. One solution was to use nn.LazyLinear. It...
## Motivation Put 'filenames' in test.py out_items to describe which file the result is. ## Checklist **Before PR**: - [x] Pre-commit or other linting tools are used to fix the...
## Motivation Support test time augmentation. Reference [mmdet tta](https://github.com/open-mmlab/mmdetection/blob/master/mmdet/datasets/pipelines/test_time_aug.py) ## Checklist **Before PR**: - [x] Pre-commit or other linting tools are used to fix the potential lint issues. - [x]...
## Motivation [Learning to Resize Images for Computer Vision Tasks](https://arxiv.org/pdf/2103.09950.pdf) ## Result CUB200 bs8×1GPU | model | top1 acc | | ---- | ---- | | swin-l | 91.6638 |...
## Motivation Support ViT Anti Oversmoothing. [Anti-Oversmoothing in Deep Vision Transformers via the Fourier Domain Analysis: From Theory to Practice](https://arxiv.org/pdf/2203.05962v1.pdf) [official repo](https://github.com/VITA-Group/ViT-Anti-Oversmoothing) Since imagenet training is difficult at hand, I...
## Motivation Support stop aug hook. This work is based on [mmdet yolox mode switch hook](https://github.com/open-mmlab/mmdetection/blob/master/mmdet/core/hook/yolox_mode_switch_hook.py). [Data Augmentation Revisited: Rethinking the Distribution Gap between Clean and Augmented Data](https://arxiv.org/pdf/1909.09148.pdf) ## Use...
## Motivation Add `MTSPPF` block and `add_ppf` param for YOLOX-PAI. [YOLOX-PAI: An Improved YOLOX, Stronger and Faster than YOLOv6](https://arxiv.org/abs/2208.13040) [official repo](https://github.com/alibaba/EasyCV/tree/master/configs/detection/yolox) ## Related PR https://github.com/open-mmlab/mmdetection/pull/8778 ## Checklist **Before PR**: -...
## Motivation [HorNet: Efficient High-Order Spatial Interactions with Recursive Gated Convolutions](https://arxiv.org/pdf/2207.14284v2.pdf) [github](https://github.com/raoyongming/HorNet) ## Weights [hornet-tiny](https://tmp.link/f/63158733df58b) [hornet-tiny-gf](https://tmp.link/f/6315876bbed10) [hornet-small](https://tmp.link/f/63158901bf8e9) [hornet-small-gf](https://tmp.link/f/6315891a51464) [hornet-base](https://tmp.link/f/6315878fd7712) [hornet-base-gf](https://tmp.link/f/631587b45f460) [hornet-large](https://tmp.link/f/63158801e8ec8) [hornet-large-gf](https://tmp.link/f/631588e615904) [hornet-large-gf384](https://tmp.link/f/63158850ea0f8) ## Checklist **Before PR**: - [x] Pre-commit...
## Motivation ### 1. Use new Architecture named series previous arch names : 'tiny', 'small', 'base' and 'large' previous arch names : 'b0', 'b1', 'b2' and 'b3' ..... And the...