[CVPR 2017] Learning a Deep Embedding Model for Zero-Shot Learning

code

需求1： 语义到视觉映射 ⇒ 视觉到语义映射

---

问题1：Can’t call numpy() on Variable that requires grad. Use var.detach().numpy() instead

分析：
kNNClassify的输入要求是numpy类型的
x_pred[i, :]是GPU上的Tensor

Traceback (most recent call last):
File “feat2attr.py”, line 119, in
print(compute_accuracy(test_x, test_att, test_att2label, test_x2label))
File “feat2attr.py”, line 24, in compute_accuracy
outputLabel = kNN.kNNClassify(x_pred[i, :], test_att, test_id, 1)
File “/home/yhr/DeepEmbeddingModel_ZSL-Pytorch/kNN.py”, line 30, in kNNClassify
diff = tile(newInput, (numSamples, 1)) - dataSet # Subtract element-wise
File “/home/yhr/anaconda2/envs/pytorch1.0/lib/python3.5/site-packages/numpy/lib/shape_base.py”, line 1147, in tile
c = _nx.array(A, copy=False, subok=True, ndmin=d)
File “/home/yhr/anaconda2/envs/pytorch1.0/lib/python3.5/site-packages/torch/tensor.py”, line 458, in array
return self.numpy()
RuntimeError: Can’t call numpy() on Variable that requires grad. Use var.detach().numpy() instead.

解决方案：

x_pred[i, :].cpu().detach().numpy()

1. cpu()将变量从GPU上转移到CPU上
2. detach.numpy()将Tensor转成numpy