FBSD
FBSD copied to clipboard
chaomaer
PyTorch Code for Feature Boosting, Suppression, and Diversification for Fine-Grained Visual Classification
When I trained witha 2 fast 1080TI,12GB graphics card, after a few epochs, the accuracy of the model dropped to 40%, and then after about 100 epochs of training, the...
您好，我在运行过程中遇到了这样的问题。通过Debug发现，在FSBM模块中，如果输入数据的width维度无法整除参数k，那么fms_softmax的第2为会变成 k+1 而不是 k。这样会导致执行代码fms_boost = fm + self.alpha*(fm * fms_softmax_boost) 时，出现维度不匹配的错误。请问这应该如何处理？
At present, there is only training code. I want to see the visual results and the judgment of images using the network
