MiZhangWhuer

How to switch between GPU and CPU?

2

In the previous caffe implementation, the GPU/CPU switching is controlled by SyncedMemory class. How can I achieve this in caffe2? Should I specify the functions, such as cudaMemcpy, in our...

Data Parallel for multi-gpu training in mergelayer branch

2

Thanks for sharing this code. I have followed your update on mergelayer branch where the recognition part is implemented. However, there is still some issues regrading to data parallel. I...

How to generate MpiTraceEvents.h header?

1

Hi, thanks for the open source project. When I trying to compile from source code using VisualStudio 2017, the solution always prompts: could not find the header #include "MpiTraceEvents.h" (mpitrace.h)...

Shape Operator support

1

When I trying to import FCN8s onnx model generated from PyTorch, the following error occurs: "op_type {} is currently not supported for NNP conversion". After checking the code, we found...

In the metrics.py, the overall accuracy is computed as: acc = np.diag(hist).sum() / hist.sum() In fact, h OA = (TP+TN)/(TP+FP+FN+TN) following the fomular in ttps://en.wikipedia.org/wiki/Confusion_matrix It is probable the TN...

space_to_depth function meaning?

3

Hi, thanks for sharing this wonderful repo. I found that there is a function space_to_depth defined as follows: def space_to_depth(input, block_size): block_size = int(block_size) block_size_sq = block_size * block_size output...

mindcv在下游任务作为backbone提取特征使用，需要新的支持

2

当前mindcv提供的backbone在分割、检测下游任务迁移，存在如下问题： 参考示例：https://github.com/mindspore-lab/models/blob/master/official/cv/OCRNet/README.md 1. 多尺度特征的获取： 针对没有skip-connection的模型，如vgg系列，backbone的多尺度特征无法获取，需要提供可获取模型特征的方法 2. 大模型缺乏特征unpatchfy机制。 transformer系列（vistformer、swin_transformer、vit等）其中，vistformer、swin_transformer仅能以固定的尺寸224、386作为特征提取器，并且不能进行多尺度特征提取；vit的提取器，除了固定尺寸、不能抽取多尺度特征的问题外，docoder部分不能剥离图片分类任务，直接将[B,C,H,W]的特征用于分割任务。 3. 所有的backbone特征，没有提供特征冻结接口，不能进行仅部分参数微调，需要全量微调，导致训练效率降低，所有下游模型模型参数都需要从头训练。

Will the MagmaDNN keep updating?

6

Hi, greate work！ I've successfully compiled the across win/linux platform. And I tried to do some modification to the current version that related to the GPU memory management issue. How...

Thanks for providing the traning source for FM model. I notice that there is the scripts **_fourm/vq/__init__.py_** to pause the pre-trained tokens. However, there is no scripts that can parse...