ruri
ruri
EMA失效
在基于Fleet开发的代码中使用EMA会失效 代码见： https://github.com/PaddlePaddle/PaddleClas/blob/master/tools/program.py#L364 https://github.com/PaddlePaddle/PaddleClas/blob/master/tools/train.py#L119  EMA参考 https://www.paddlepaddle.org.cn/documentation/docs/zh/develop/api_cn/optimizer_cn/ExponentialMovingAverage_cn.html#exponentialmovingaverage 怀疑是distributed_optimizer的问题....
After testing with_mask in default demo_lazy.py There are jagged edges around the body that cannot be ignored, Is it not good enough? or should I adjust some args to improve,...
when I tested demo ```150_16_dinat_l_oneformer_coco_100ep.pth``` with related config ```configs/coco/dinat/oneformer_dinat_large_bs16_100ep.yaml``` I got a ```natten``` error ```Input axes must be greater than or equal to the product of kernel size and dilation....
- [x] add ImageNet benchmark script to test speed. Feed Data  No Feed Data 
Speedup Image Classification training speed --- - [x] add eval shape arguments, for different image shape processing, like inception v4 - [x] add multiprocess reader, set cv2 as the default...
- add CE for image classification How to enable ce testing? set ```--enable_ce=True```
- add TSN SeResNeXt model in PaddleVideo
- reduce reader_buffer size of xmap_readers in image classification 2048->64 - pass args:warm_up_epochs in func: cosine_warmup_decay
fix bug