search

zkkli / I-ViT

[ICCV 2023] I-ViT: Integer-only Quantization for Efficient Vision Transformer Inference
Apache License 2.0
148 stars 13 forks source link

您好，我这里所有图片的输出为什么都是这样的 #4

Open husnxjsjsb opened 9 months ago

husnxjsjsb commented 9 months ago

lsy

husnxjsjsb commented 9 months ago

【801 509 71 377 149】

zkkli commented 9 months ago

Hello，得到的输出是：Top-5的分类标签。

husnxjsjsb commented 9 months ago

感谢您的回答，请问我应该怎么从这个输出的结果判断出预测的类别呢，我没有找到可以将输出的标签和我想得到的预测结果相联系的方法。

zouzihan3 commented 9 months ago

我记得如果使用ImageNet的话，有解析方法的。其中801可能就对应了imagenet的一个分类，你可以参考一下如下博客。 https://blog.csdn.net/weixin_47317192/article/details/126629218

dedoogong commented 1 month ago

if your image was the default cat, the result is totatlly wrong as I faced. I don't know why the converted model and TVM run gives totally wrong result. 【801 509 71 377 149】itself shows that the predicted top 5 classes has no similar looking.