Python Error 汇总

ModuleNotFoundError

1.No module named ‘pycocotools‘

升级为2.0.1地址
cocoapi兼容window10：代码地址
下载好了之后通过cmd进入到pythonAPI下
运行一下代码

# install pycocotools locally
python setup.py build_ext --inplace
 
# install pycocotools to the Python site-packages
python setup.py build_ext install

安装完成成功import pycocotools

TypeError

1.can‘t convert cuda:0 device type tensor to numpy

原因分析：
CUDA tensor格式的数据改成numpy时，需要先将其转换成cpu float-tensor随后再转到numpy格式。
因为numpy不能读取CUDA tensor ，需要将它转化为 CPU tensor

(loss / (iteration + 1)).numpy()

改为

(loss / (iteration + 1)).cpu().detach().numpy()

2.Object of type ‘ndarray’ is not JSON serializable

json不认numpy的array，全部改为str类型

3.TypeError: src data type = 17 is not supported

在cvtColor前面加上np.array。
img_np = np.array(img,np.uint8)
frame = cv2.cvtColor(img_np, cv2.COLOR_BGR2RGB)

RuntimeError

1.CUDA out of memory. Tried to allocate 128.00 MiB (GPU 0； 2.00 GiB total capacity； 1

解决链接:---->

2.RuntimeError: cuDNN error: CUDNN_STATUS_INTERNAL_ERROR

解决办法：调小batch-size试试

3.RuntimeError: Expected object of scalar type Long but got scalar type Int for argument #2 ‘target’ in call to thnn_nll

解决方式：发现是关于label的错误,将loss = loss_function(out, label) 变为loss = loss_function(out, label.long())

4.RuntimeError: Sizes of tensors must match except in dimension 2. Got 28 and 27 (The offending index is 0)

原因：尺寸不匹配问题
解决：定位到DW卷积处，发现卷积核为5，而padding为1，不满足(k-1)/2=p,所以这里的padding应该为2，问题解决。

5.RuntimeError: “log_cuda” not implemented for ‘Long’

原因：Long类型的数据不支持log对数运算, 为什么Tensor是Long类型? 因为创建numpy 数组时没有指定dtype, 默认使用的是int64, 所以从numpy array转成torch.Tensor后, 数据类型变成了Long

loss = torch.where(true == 1, (1 - pred) ** gamma * (torch.log(pred)), pred ** gamma * (torch.log(1 - pred)))

改为

loss = torch.where(true == 1, (1 - pred) ** gamma * (torch.log(torch.from_numpy(pred))), pred ** gamma * (torch.log(torch.from_numpy(1 - pred))))

编译问题

1.windows编译detectron2

出现问题error: command ‘C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.1\bin\nvcc.exe’ failed with exit status 1
解决：下载https://github.com/conansherry/detectron2替换了/detectron2/layers/csrc/nms_rotated/目录下的3个文件，并成功了。

2.windows编译PSROIAlign报错

• error: command ‘C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v10.0\bin\nvcc.exe’ failed with exit status 1

• 

• 原因:出现这个错误的原因是，数据类型不对，无法编译通过。

• 解决方式:根据红色路径找到这个cu文件，在进行修改()

// line 275
dim3 grid(std::min(THCCeilDiv((long)output_size, 512L), 4096L));
// line 320
dim3 grid(std::min(THCCeilDiv((long)grad.numel(), 512L), 4096L));

实际情况如下

• 

3.RuntimeError: view size is not compatible with input tensor‘s size and stride

这是因为view()需要Tensor中的元素地址是连续的，但可能出现Tensor不连续的情况，所以先用 .contiguous() 将其在内存中变成连续分布：改为

OnnxError

Unexpected input data type. Actual: (N11onnxruntime17PrimitiveDataTypeIdEE)

def process_image(img):                                                                            
    img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)                                                     
    h, w = img.shape[:2]                                                                           
    scale_x = float(input_shape[1]) / w                                                            
    scale_y = float(input_shape[0]) / h                                                            
    img = cv2.resize(img, None, None, fx=scale_x, fy=scale_y, interpolation=cv2.INTER_CUBIC)       
    return np.expand_dims(img, axis=0).astype(np.float32)                                          
                                                                                                   
def predict(image, shape):                                                                         
    # image1=np.asarray(image)                                                                     
    # print(image1)                                                                                
    input_data = np.random.random((1, 416, 416, 3)).astype('float32')                              
    # print(input_data)                                                                            
    # print('1',type(input_data))                                                                  
    # print('2',type(image1.shape))                                                                
    Yolov4_onnx(input_data=image)                                                                  

将源代码中的np.expand_dims(img, axis=0)改为np.expand_dims(img, axis=0).astype(np.float32)

BrokenPipeError

1.Broken pipe

• 错误原因 多线程问题
• 将num_workers的值改为0，代表不适用多线程

AssertionError

1.Results do not correspond to current coco set

AttributeError

1.AttributeError: ‘tuple‘ object has no attribute ‘cuda‘

原因：target是tuple类型，但.conda()需要是tensor类型
解决：tuple——np.array——tensor（中间需要np.array中转；且np.array的元素需要是int或float（原本是str），使用.astype(int)转化即可。
具体代码：

        target=np.array(target).astype(int)
        target=torch.from_numpy(target)
        target = target.cuda()

ValueError

ValueError: matrix contains invalid numeric entries

解决办法：在_lasp.py函数中，加入一下判断语句，过滤掉这个错误。

    if np.any(np.isneginf(cost_matrix) | np.isnan(cost_matrix)):
        raise ValueError("matrix contains invalid numeric entries")

Other

QObject::moveToThread: Current thread (0x562cd93b0cd0) is not the object’s thread (0x562cd9528330). Cannot move to target thread (0x562cd93b0cd0)
pip uninstall opencv-python
pip install opencv-python==4.1.1.26 解决

size mismatch for stage4.3.branch2.3.weight: copying a param with shape torch.Size([96, 1, 3, 3]) from checkpoint, the shape in current model is torch.Size([96, 1, 5, 5]).
原因:由于stage模块里面的参数是有差别的，所以需要把预训练模型的stage模块进行去除
解决：

        if pretrain:
            url = model_urls['shufflenetv2_{}'.format(self.model_size)]
            if url is not None:
                pretrained_state_dict = model_zoo.load_url(url)
                print('=> loading pretrained model {}'.format(url))
                model_dict = self.state_dict()
                # 重新制作预训练的权重，主要是减去参数不匹配的层，楼主这边层名为“fc”
                pretrained_dict = {k: v for k, v in pretrained_state_dict.items() if (k in model_dict and 'stage' not in k)}
                # 更新权重
                model_dict.update(pretrained_dict)

                self.load_state_dict(model_dict, strict=False)

CUDA_ERROR_OUT_OF_MEMORY: out of memory

config = tf.ConfigProto(allow_soft_placement=True, log_device_placement=False)
config.gpu_options.allow_growth = True
session = tf.Session(config=config)

Pycharm无法打开问题

pytorch计算模型参数量报错： size mismatch for stage2.0.branch2.3.weight: copying a param with shape torch.Size([58, 1, 3, 3]) from checkpoint, the shape in current model is torch.Size([58, 1, 5, 5]).

在加载模型时加上上面这句话，因为是多GPU训练出来的模型。

    model = shufflenet_v2_x1_0(num_classes=args.num_classes).to(device)

    if device !='cpu':
        model = torch.nn.DataParallel(model)
        cudnn.benchmark = True
        model = model.cuda()
    if args.weights != "":
        if os.path.exists(args.weights):
            weights_dict = torch.load(args.weights, map_location=device)
            model_dict = model.state_dict()
            #load_weights_dict = {k: v for k, v in weights_dict.items()  if model.state_dict()[k].numel() == v.numel()}
            load_weights_dict = {k: v for k, v in weights_dict.items() if (k in model_dict and 'fc' not in k)}

            model_dict.update(load_weights_dict)

            print(model.load_state_dict(model_dict, strict=False))
        else:
            raise FileNotFoundError("not found weights file: {}".format(args.weights))