lua中的分段错误（核心已转储）

Question

我正在尝试运行此代码https://github.com/Winfrand/C-MIL，并且得到了此输出 分段错误（核心已转储）

当我尝试调试代码时，我发现代码在执行时就停止了

local output = self.model:forward(inputs) 

功能

Optim:optimize(optimMethod, inputs, targets, criterion, scale) 

在

fbnn_Optim.lua文件，第145行

当我分别打印模型和输入的内容时

print(model)
nn.Sequential {[input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
(1): nn.ParallelTable {
input
  |`-> (1): nn.Sequential {
  |      [input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> (7) -> (8) -> (9) -> (10) -> (11) -> (12) -> (13) -> (14) -> output]
  |      (1): cudnn.SpatialConvolution(3 -> 64, 11x11, 4,4)
  |      (2): cudnn.ReLU
  |      (3): cudnn.SpatialCrossMapLRN
  |      (4): cudnn.SpatialMaxPooling(3x3, 2,2)
  |      (5): cudnn.SpatialConvolution(64 -> 256, 5x5, 1,1, 2,2)
  |      (6): cudnn.ReLU
  |      (7): cudnn.SpatialCrossMapLRN
  |      (8): cudnn.SpatialMaxPooling(3x3, 2,2)
  |      (9): cudnn.SpatialConvolution(256 -> 256, 3x3, 1,1, 1,1)
  |      (10): cudnn.ReLU
  |      (11): cudnn.SpatialConvolution(256 -> 256, 3x3, 1,1, 1,1)
  |      (12): cudnn.ReLU
  |      (13): cudnn.SpatialConvolution(256 -> 256, 3x3, 1,1, 1,1)
  |      (14): cudnn.ReLU
  |    }
   `-> (2): nn.Identity
   ... -> output
}
(2): nn.ConcatTable {
input
  |`-> (1): nn.Sequential {
  |      [input -> (1) -> (2) -> (3) -> output]
  |      (1): RectangularRingRoiPooling
  |      (2): nn.View(-1)
  |      (3): nn.Sequential {
  |        [input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
  |        (1): nn.Linear(9216 -> 4096)
  |        (2): cudnn.ReLU
  |        (3): nn.Dropout(0.500000)
  |        (4): nn.Linear(4096 -> 4096)
  |        (5): cudnn.ReLU
  |        (6): nn.Dropout(0.500000)
  |      }
  |    }
  |`-> (2): nn.Sequential {
  |      [input -> (1) -> (2) -> (3) -> output]
  |      (1): RectangularRingRoiPooling
  |      (2): nn.View(-1)
  |      (3): nn.Sequential {
  |        [input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
  |        (1): nn.Linear(9216 -> 4096)
  |        (2): cudnn.ReLU
  |        (3): nn.Dropout(0.500000)
  |        (4): nn.Linear(4096 -> 4096)
  |        (5): cudnn.ReLU
  |        (6): nn.Dropout(0.500000)
  |      }
  |    }
   `-> (3): nn.Sequential {
         [input -> (1) -> (2) -> (3) -> output]
         (1): RectangularRingRoiPooling
         (2): nn.View(-1)
         (3): nn.Sequential {
           [input -> (1) -> (2) -> (3) -> (4) -> (5) -> (6) -> output]
           (1): nn.Linear(9216 -> 4096)
           (2): cudnn.ReLU
           (3): nn.Dropout(0.500000)
           (4): nn.Linear(4096 -> 4096)
           (5): cudnn.ReLU
           (6): nn.Dropout(0.500000)
         }
       }
   ... -> output
} 
(3): nn.ConcatTable {
input
  |`-> (1): nn.Identity
  |`-> (2): nn.Sequential {
  |      [input -> (1) -> (2) -> (3) -> output]
  |      (1): nn.SelectTable(1)
  |      (2): nn.Linear(4096 -> 21)
  |      (3): cudnn.SpatialSoftMax
  |    }
   `-> (3): nn.Sequential {
         [input -> (1) -> (2) -> (3) -> output]
         (1): nn.SelectTable(1)
         (2): nn.Linear(4096 -> 21)
         (3): cudnn.SpatialSoftMax
       }
   ... -> output
}
(4): nn.ReWeight
(5): nn.ParallelTable {
input
  |`-> (1): nn.Sequential {
  |      [input -> (1) -> (2) -> output]
  |      (1): nn.ConcatTable {
  |        input
  |          |`-> (1): nn.Sequential {
  |          |      [input -> (1) -> (2) -> (3) -> (4) -> output]
  |          |      (1): nn.SelectTable(1)
  |          |      (2): nn.Linear(4096 -> 20)
  |          |      (3): nn.View(-1)
  |          |      (4): nn.Squeeze
  |          |    }
  |           `-> (2): nn.Sequential {
  |                 [input -> (1) -> (2) -> (3) -> output]
  |                 (1): nn.ConcatTable {
  |                   input
  |                     |`-> (1): nn.Sequential {
  |                     |      [input -> (1) -> (2) -> output]
  |                     |      (1): nn.SelectTable(2)
  |                     |      (2): nn.Linear(4096 -> 20)
  |                     |    }
  |                      `-> (2): nn.Sequential {
  |                            [input -> (1) -> (2) -> (3) -> output]
  |                            (1): nn.SelectTable(3)
  |                            (2): nn.Linear(4096 -> 20)
  |                            (3): nn.MulConstant
  |                          }
  |                      ... -> output
  |                 }
  |                 (2): nn.CAddTable
  |                 (3): nn.View(-1)
  |               }
  |           ... -> output
  |      }
  |      (2): nn.ContinuationSubset
  |    }
  |`-> (2): nn.Identity
   `-> (3): nn.Identity
   ... -> output
}
(6): nn.ContinuationDetector
}

print(inputs)
{
1 : CudaTensor - size: 1x3x608x586
2 : FloatTensor - size: 1x1963x5
}

这是此功能的代码

function Optim:optimize(optimMethod, inputs, targets, criterion, scale)
assert(optimMethod)
assert(inputs)
assert(targets)
assert(criterion)
assert(self.modulesToOptState)

self.model:zeroGradParameters()

local output = self.model:forward(inputs)

if type(targets) == 'table' then
    if type(batch_box_labels_gpu) == 'table' then
        for i=1,#batch_box_labels_gpu do
            targets[i+1] = batch_box_labels_gpu[i]
        end
    else
        targets[2] = batch_box_labels_gpu
    end
end

print(output)
print(targets)
local err = criterion:forward(output, targets)

if err ~= 0 then

local df_do = criterion:backward(output, targets)
self.model:backward(inputs, df_do, scale)

-- We'll set these in the loop that iterates over each module. Get them
-- out here to be captured.
local curGrad
local curParam
local function fEvalMod(x)
    return err, curGrad
end

for curMod, opt in pairs(self.modulesToOptState) do
    on_device_for_module(curMod, function()
        local curModParams = self.weight_bias_parameters(curMod)
        -- expects either an empty table or 2 element table, one for weights
        -- and one for biases
        assert(pl.tablex.size(curModParams) == 0 or
               pl.tablex.size(curModParams) == 2)
        if curModParams then
            for i, tensor in ipairs(curModParams) do
                if curModParams[i] then
                    -- expect param, gradParam pair
                    curParam, curGrad = table.unpack(curModParams[i])
                    assert(curParam and curGrad)
                    optimMethod(fEvalMod, curParam, opt[i])
                end
            end
        end
    end)
end

end
return err, output
end

因此，如果有人可以指导我解决此问题，我将永远感激

提前谢谢