The problem of classification models

[lindadamama]

Hello, I just want to perform a two-classification result task, such as the forward and reverse directions of the object, and the ng and ok of the object. Can I do it?

dotnet/TorchSharp#1463

There is another question. Torchsharp admits that there will be problems when the model executes EVAL() ,Have you ever encountered the same problem with this model saving method?

The third question is, why are the numbers in the following file decimals? Shouldn't the values ​​of X and Y corresponding to the coordinates be given?
https://github.com/IntptrMax/YoloSharp/blob/master/YoloSharpDemo/Assets/DataSets/coco128-seg/labels/train2017/000000000030.txt

[IntptrMax]

Yolo can do that. But if you just want to do two-classification task, and don't care about where is the object to detect, ResNet can do better.
https://github.com/IntptrMax/ResnetTrain

[IntptrMax]

1. Yolo-CLS has been added. Please have a try.
2. EVAL is not completely fit for Yolosharp now, because of the shapes of the tensors are not reshaped for eval, it will be fixed with export.
3. Label is in coco type: ID CX/imageWidth CY/imageHeight W/imageWidth H/imageHeight

[IntptrMax]

Hello, I just want to perform a two-classification result task, such as the forward and reverse directions of the object, and the ng and ok of the object. Can I do it?

dotnet/TorchSharp#1463

There is another question. Torchsharp admits that there will be problems when the model executes EVAL() ,Have you ever encountered the same problem with this model saving method?

The third question is, why are the numbers in the following file decimals? Shouldn't the values ​​of X and Y corresponding to the coordinates be given? https://github.com/IntptrMax/YoloSharp/blob/master/YoloSharpDemo/Assets/DataSets/coco128-seg/labels/train2017/000000000030.txt

I have add yolo-cls task, which maybe help.