pytorch網絡轉libtorch常見問題

### <span id="1">一、All inputs of range must be ints, found Tensor in argument 0:</span>
![](https://img2020.cnblogs.com/blog/1188959/202111/1188959-20211115105734283-880018543.png)

**問題**
<font color="red">參數類型不正確，函數的默認參數是tensor</font>

**解決措施**
<font color="red">函數傳入參數不是tensor需要注明類型</font>
我的問題是傳入參數`npoint`是一個`int`類型，沒有注明會報錯，更改如下：
由
```python
def test(npoint):
  ...
```
更改為
```python
def test(npoint: int):
  ...
```

二、Sliced expression not yet supported for subscripted assignment. File a bug if you want this:

問題
不支持賦值給切片表達式

解決措施
根據自己需求，進行修改，可利用循環替代

我將view_shape[1:] = [1] * (len(view_shape) - 1)更改為

    for i in range(1, len(view_shape)):
        view_shape[i] = 1

---

三、Tried to access nonexistent attribute or method 'len' of type 'torch.torch.nn.modules.container.ModuleList'. Did you forget to initialize an attribute in init()?

問題
forward函數中好像不支持len(nn.ModuleList())和下標訪問

解決措施
如果是一個ModuleList()可以用enumerate函數，多個同維度的可以用zip函數

我這里有兩個ModuleList()，所以采用zip函數，更改如下：
由

   for i, conv in enumerate(self.mlp_convs):
      bn = self.mlp_bns[i]
      new_points = F.relu(bn(conv(new_points)))

更改為

    for conv, bn in zip(self.mlp_convs, self.mlp_bns):
        new_points = F.relu(bn(conv(new_points)))

ref: https://github.com/pytorch/pytorch/issues/16123

---

四、Expected integer literal for index

問題和解決方法類似第三個

---

五、Arguments for call are not valid. The following variants are available

Expected a value of type 'List[Tensor]' for argument 'indices' but instead found type 'List[Optional[Tensor]]'

問題
賦值類型不對，需求是tensor，但給的是int

解決措施

• 方法1
  將int類型的數N用torch.tensor(N)代替
  由

mask = sqrdists > radius ** 2
group_idx[mask] = N

變為

mask = sqrdists > radius ** 2
group_idx[mask] = torch.tensor(N)

• 方法2 （速度較慢）
  用for循環替代`
  由

mask = sqrdists > radius ** 2
group_idx[mask] = N

變為

B, rows, cols = sqrdists.shape
ref_redius = radius ** 2
for b in range(B):
    for r in range(rows):
        print("r: ", r)
        for c in range(cols):
            if sqrdists[b][r][c] > ref_redius:
                group_idx[b][r][c] = N