[onni] Segmentation fault for GoogLeNet

Question

[onni] Segmentation fault for GoogLeNet

Closed this issue 6 years ago · 3 comments

HKLee2040 commented 6 years ago

When I try onni on googlenet case by following command,
onni ~/onnx_model_zoo/bvlc_googlenet/model.onnx ~/onnx_model_zoo/bvlc_googlenet/test_data_set_0/input_0.pb

The system shows me a message "Segmentation fault".
Can you help me what's going on?

[v3] Operator | Count
[v3] ------------+------
[v3] Softmax | 1
[v3] Gemm | 1
[v3] Reshape | 1
[v3] Concat | 9
[v3] Conv | 57
[v3] MaxPool | 13
[v3] Relu | 57
[v3] AveragePool | 1
[v3] LRN | 2
[v3] ------------+------
[v3] Total | 142
[v1] weight memory: 27994224
[v1] internal memory: 6422528
[v2] %conv1/7x7_s2_1[1, 64, 112, 112] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [7, 7], pads: [3, 3, 3, 3], strides: [2, 2]>(%data_0[1, 3, 224, 224], %conv1/7x7_s2_w_0[64, 3, 7, 7], %conv1/7x7_s2_b_0[64])
[v2] %conv1/7x7_s2_1[1, 64, 112, 112] = Relu(%conv1/7x7_s2_1[1, 64, 112, 112])
[v2] %pool1/3x3_s2_1[1, 64, 56, 56] = MaxPool<auto_pad: "NOTSET", kernel_shape: [3, 3], pads: [0, 0, 2, 2], storage_order: 0, strides: [2, 2]>(%conv1/7x7_s2_1[1, 64, 112, 112])
[v2] %pool1/norm1_1[1, 64, 56, 56] = LRN<alpha: 0.0001, beta: 0.75, bias: 1, size: 5>(%pool1/3x3_s2_1[1, 64, 56, 56])
[v2] %conv2/3x3_reduce_1[1, 64, 56, 56] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%pool1/norm1_1[1, 64, 56, 56], %conv2/3x3_reduce_w_0[64, 64, 1, 1], %conv2/3x3_reduce_b_0[64])
[v2] %conv2/3x3_reduce_1[1, 64, 56, 56] = Relu(%conv2/3x3_reduce_1[1, 64, 56, 56])
[v2] %conv2/3x3_1[1, 192, 56, 56] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [3, 3], pads: [1, 1, 1, 1], strides: [1, 1]>(%conv2/3x3_reduce_1[1, 64, 56, 56], %conv2/3x3_w_0[192, 64, 3, 3], %conv2/3x3_b_0[192])
[v2] %conv2/3x3_1[1, 192, 56, 56] = Relu(%conv2/3x3_1[1, 192, 56, 56])
[v2] %conv2/norm2_1[1, 192, 56, 56] = LRN<alpha: 0.0001, beta: 0.75, bias: 1, size: 5>(%conv2/3x3_1[1, 192, 56, 56])
[v2] %pool2/3x3_s2_1[1, 192, 28, 28] = MaxPool<auto_pad: "NOTSET", kernel_shape: [3, 3], pads: [0, 0, 2, 2], storage_order: 0, strides: [2, 2]>(%conv2/norm2_1[1, 192, 56, 56])
[v2] %inception_3a/1x1_1[1, 64, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%pool2/3x3_s2_1[1, 192, 28, 28], %inception_3a/1x1_w_0[64, 192, 1, 1], %inception_3a/1x1_b_0[64])
[v2] %inception_3a/1x1_1[1, 64, 28, 28] = Relu(%inception_3a/1x1_1[1, 64, 28, 28])
[v2] %inception_3a/3x3_reduce_1[1, 96, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%pool2/3x3_s2_1[1, 192, 28, 28], %inception_3a/3x3_reduce_w_0[96, 192, 1, 1], %inception_3a/3x3_reduce_b_0[96])
[v2] %inception_3a/3x3_reduce_1[1, 96, 28, 28] = Relu(%inception_3a/3x3_reduce_1[1, 96, 28, 28])
[v2] %inception_3a/3x3_1[1, 128, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [3, 3], pads: [1, 1, 1, 1], strides: [1, 1]>(%inception_3a/3x3_reduce_1[1, 96, 28, 28], %inception_3a/3x3_w_0[128, 96, 3, 3], %inception_3a/3x3_b_0[128])
[v2] %inception_3a/3x3_1[1, 128, 28, 28] = Relu(%inception_3a/3x3_1[1, 128, 28, 28])
[v2] %inception_3a/5x5_reduce_1[1, 16, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%pool2/3x3_s2_1[1, 192, 28, 28], %inception_3a/5x5_reduce_w_0[16, 192, 1, 1], %inception_3a/5x5_reduce_b_0[16])
[v2] %inception_3a/5x5_reduce_1[1, 16, 28, 28] = Relu(%inception_3a/5x5_reduce_1[1, 16, 28, 28])
[v2] %inception_3a/5x5_1[1, 32, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [5, 5], pads: [2, 2, 2, 2], strides: [1, 1]>(%inception_3a/5x5_reduce_1[1, 16, 28, 28], %inception_3a/5x5_w_0[32, 16, 5, 5], %inception_3a/5x5_b_0[32])
[v2] %inception_3a/5x5_1[1, 32, 28, 28] = Relu(%inception_3a/5x5_1[1, 32, 28, 28])
[v2] %inception_3a/pool_1[1, 192, 28, 28] = MaxPool<auto_pad: "NOTSET", kernel_shape: [3, 3], pads: [1, 1, 1, 1], storage_order: 0, strides: [1, 1]>(%pool2/3x3_s2_1[1, 192, 28, 28])
[v2] %inception_3a/pool_proj_1[1, 32, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_3a/pool_1[1, 192, 28, 28], %inception_3a/pool_proj_w_0[32, 192, 1, 1], %inception_3a/pool_proj_b_0[32])
[v2] %inception_3a/pool_proj_1[1, 32, 28, 28] = Relu(%inception_3a/pool_proj_1[1, 32, 28, 28])
[v2] %inception_3a/output_1[1, 256, 28, 28] = Concat<axis: 1>(%inception_3a/1x1_1[1, 64, 28, 28], %inception_3a/3x3_1[1, 128, 28, 28], %inception_3a/5x5_1[1, 32, 28, 28], %inception_3a/pool_proj_1[1, 32, 28, 28])
[v2] %inception_3b/1x1_1[1, 128, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_3a/output_1[1, 256, 28, 28], %inception_3b/1x1_w_0[128, 256, 1, 1], %inception_3b/1x1_b_0[128])
[v2] %inception_3b/1x1_1[1, 128, 28, 28] = Relu(%inception_3b/1x1_1[1, 128, 28, 28])
[v2] %inception_3b/3x3_reduce_1[1, 128, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_3a/output_1[1, 256, 28, 28], %inception_3b/3x3_reduce_w_0[128, 256, 1, 1], %inception_3b/3x3_reduce_b_0[128])
[v2] %inception_3b/3x3_reduce_1[1, 128, 28, 28] = Relu(%inception_3b/3x3_reduce_1[1, 128, 28, 28])
[v2] %inception_3b/3x3_1[1, 192, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [3, 3], pads: [1, 1, 1, 1], strides: [1, 1]>(%inception_3b/3x3_reduce_1[1, 128, 28, 28], %inception_3b/3x3_w_0[192, 128, 3, 3], %inception_3b/3x3_b_0[192])
[v2] %inception_3b/3x3_1[1, 192, 28, 28] = Relu(%inception_3b/3x3_1[1, 192, 28, 28])
[v2] %inception_3b/5x5_reduce_1[1, 32, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_3a/output_1[1, 256, 28, 28], %inception_3b/5x5_reduce_w_0[32, 256, 1, 1], %inception_3b/5x5_reduce_b_0[32])
[v2] %inception_3b/5x5_reduce_1[1, 32, 28, 28] = Relu(%inception_3b/5x5_reduce_1[1, 32, 28, 28])
[v2] %inception_3b/5x5_1[1, 96, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [5, 5], pads: [2, 2, 2, 2], strides: [1, 1]>(%inception_3b/5x5_reduce_1[1, 32, 28, 28], %inception_3b/5x5_w_0[96, 32, 5, 5], %inception_3b/5x5_b_0[96])
[v2] %inception_3b/5x5_1[1, 96, 28, 28] = Relu(%inception_3b/5x5_1[1, 96, 28, 28])
[v2] %inception_3b/pool_1[1, 256, 28, 28] = MaxPool<auto_pad: "NOTSET", kernel_shape: [3, 3], pads: [1, 1, 1, 1], storage_order: 0, strides: [1, 1]>(%inception_3a/output_1[1, 256, 28, 28])
[v2] %inception_3b/pool_proj_1[1, 64, 28, 28] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_3b/pool_1[1, 256, 28, 28], %inception_3b/pool_proj_w_0[64, 256, 1, 1], %inception_3b/pool_proj_b_0[64])
[v2] %inception_3b/pool_proj_1[1, 64, 28, 28] = Relu(%inception_3b/pool_proj_1[1, 64, 28, 28])
[v2] %inception_3b/output_1[1, 480, 28, 28] = Concat<axis: 1>(%inception_3b/1x1_1[1, 128, 28, 28], %inception_3b/3x3_1[1, 192, 28, 28], %inception_3b/5x5_1[1, 96, 28, 28], %inception_3b/pool_proj_1[1, 64, 28, 28])
[v2] %pool3/3x3_s2_1[1, 480, 14, 14] = MaxPool<auto_pad: "NOTSET", kernel_shape: [3, 3], pads: [0, 0, 2, 2], storage_order: 0, strides: [2, 2]>(%inception_3b/output_1[1, 480, 28, 28])
[v2] %inception_4a/1x1_1[1, 192, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%pool3/3x3_s2_1[1, 480, 14, 14], %inception_4a/1x1_w_0[192, 480, 1, 1], %inception_4a/1x1_b_0[192])
[v2] %inception_4a/1x1_1[1, 192, 14, 14] = Relu(%inception_4a/1x1_1[1, 192, 14, 14])
[v2] %inception_4a/3x3_reduce_1[1, 96, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%pool3/3x3_s2_1[1, 480, 14, 14], %inception_4a/3x3_reduce_w_0[96, 480, 1, 1], %inception_4a/3x3_reduce_b_0[96])
[v2] %inception_4a/3x3_reduce_1[1, 96, 14, 14] = Relu(%inception_4a/3x3_reduce_1[1, 96, 14, 14])
[v2] %inception_4a/3x3_1[1, 208, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [3, 3], pads: [1, 1, 1, 1], strides: [1, 1]>(%inception_4a/3x3_reduce_1[1, 96, 14, 14], %inception_4a/3x3_w_0[208, 96, 3, 3], %inception_4a/3x3_b_0[208])
[v2] %inception_4a/3x3_1[1, 208, 14, 14] = Relu(%inception_4a/3x3_1[1, 208, 14, 14])
[v2] %inception_4a/5x5_reduce_1[1, 16, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%pool3/3x3_s2_1[1, 480, 14, 14], %inception_4a/5x5_reduce_w_0[16, 480, 1, 1], %inception_4a/5x5_reduce_b_0[16])
[v2] %inception_4a/5x5_reduce_1[1, 16, 14, 14] = Relu(%inception_4a/5x5_reduce_1[1, 16, 14, 14])
[v2] %inception_4a/5x5_1[1, 48, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [5, 5], pads: [2, 2, 2, 2], strides: [1, 1]>(%inception_4a/5x5_reduce_1[1, 16, 14, 14], %inception_4a/5x5_w_0[48, 16, 5, 5], %inception_4a/5x5_b_0[48])
[v2] %inception_4a/5x5_1[1, 48, 14, 14] = Relu(%inception_4a/5x5_1[1, 48, 14, 14])
[v2] %inception_4a/pool_1[1, 480, 14, 14] = MaxPool<auto_pad: "NOTSET", kernel_shape: [3, 3], pads: [1, 1, 1, 1], storage_order: 0, strides: [1, 1]>(%pool3/3x3_s2_1[1, 480, 14, 14])
[v2] %inception_4a/pool_proj_1[1, 64, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4a/pool_1[1, 480, 14, 14], %inception_4a/pool_proj_w_0[64, 480, 1, 1], %inception_4a/pool_proj_b_0[64])
[v2] %inception_4a/pool_proj_1[1, 64, 14, 14] = Relu(%inception_4a/pool_proj_1[1, 64, 14, 14])
[v2] %inception_4a/output_1[1, 512, 14, 14] = Concat<axis: 1>(%inception_4a/1x1_1[1, 192, 14, 14], %inception_4a/3x3_1[1, 208, 14, 14], %inception_4a/5x5_1[1, 48, 14, 14], %inception_4a/pool_proj_1[1, 64, 14, 14])
[v2] %inception_4b/1x1_1[1, 160, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4a/output_1[1, 512, 14, 14], %inception_4b/1x1_w_0[160, 512, 1, 1], %inception_4b/1x1_b_0[160])
[v2] %inception_4b/1x1_1[1, 160, 14, 14] = Relu(%inception_4b/1x1_1[1, 160, 14, 14])
[v2] %inception_4b/3x3_reduce_1[1, 112, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4a/output_1[1, 512, 14, 14], %inception_4b/3x3_reduce_w_0[112, 512, 1, 1], %inception_4b/3x3_reduce_b_0[112])
[v2] %inception_4b/3x3_reduce_1[1, 112, 14, 14] = Relu(%inception_4b/3x3_reduce_1[1, 112, 14, 14])
[v2] %inception_4b/3x3_1[1, 224, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [3, 3], pads: [1, 1, 1, 1], strides: [1, 1]>(%inception_4b/3x3_reduce_1[1, 112, 14, 14], %inception_4b/3x3_w_0[224, 112, 3, 3], %inception_4b/3x3_b_0[224])
[v2] %inception_4b/3x3_1[1, 224, 14, 14] = Relu(%inception_4b/3x3_1[1, 224, 14, 14])
[v2] %inception_4b/5x5_reduce_1[1, 24, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4a/output_1[1, 512, 14, 14], %inception_4b/5x5_reduce_w_0[24, 512, 1, 1], %inception_4b/5x5_reduce_b_0[24])
[v2] %inception_4b/5x5_reduce_1[1, 24, 14, 14] = Relu(%inception_4b/5x5_reduce_1[1, 24, 14, 14])
[v2] %inception_4b/5x5_1[1, 64, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [5, 5], pads: [2, 2, 2, 2], strides: [1, 1]>(%inception_4b/5x5_reduce_1[1, 24, 14, 14], %inception_4b/5x5_w_0[64, 24, 5, 5], %inception_4b/5x5_b_0[64])
[v2] %inception_4b/5x5_1[1, 64, 14, 14] = Relu(%inception_4b/5x5_1[1, 64, 14, 14])
[v2] %inception_4b/pool_1[1, 512, 14, 14] = MaxPool<auto_pad: "NOTSET", kernel_shape: [3, 3], pads: [1, 1, 1, 1], storage_order: 0, strides: [1, 1]>(%inception_4a/output_1[1, 512, 14, 14])
[v2] %inception_4b/pool_proj_1[1, 64, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4b/pool_1[1, 512, 14, 14], %inception_4b/pool_proj_w_0[64, 512, 1, 1], %inception_4b/pool_proj_b_0[64])
[v2] %inception_4b/pool_proj_1[1, 64, 14, 14] = Relu(%inception_4b/pool_proj_1[1, 64, 14, 14])
[v2] %inception_4b/output_1[1, 512, 14, 14] = Concat<axis: 1>(%inception_4b/1x1_1[1, 160, 14, 14], %inception_4b/3x3_1[1, 224, 14, 14], %inception_4b/5x5_1[1, 64, 14, 14], %inception_4b/pool_proj_1[1, 64, 14, 14])
[v2] %inception_4c/1x1_1[1, 128, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4b/output_1[1, 512, 14, 14], %inception_4c/1x1_w_0[128, 512, 1, 1], %inception_4c/1x1_b_0[128])
[v2] %inception_4c/1x1_1[1, 128, 14, 14] = Relu(%inception_4c/1x1_1[1, 128, 14, 14])
[v2] %inception_4c/3x3_reduce_1[1, 128, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4b/output_1[1, 512, 14, 14], %inception_4c/3x3_reduce_w_0[128, 512, 1, 1], %inception_4c/3x3_reduce_b_0[128])
[v2] %inception_4c/3x3_reduce_1[1, 128, 14, 14] = Relu(%inception_4c/3x3_reduce_1[1, 128, 14, 14])
[v2] %inception_4c/3x3_1[1, 256, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [3, 3], pads: [1, 1, 1, 1], strides: [1, 1]>(%inception_4c/3x3_reduce_1[1, 128, 14, 14], %inception_4c/3x3_w_0[256, 128, 3, 3], %inception_4c/3x3_b_0[256])
[v2] %inception_4c/3x3_1[1, 256, 14, 14] = Relu(%inception_4c/3x3_1[1, 256, 14, 14])
[v2] %inception_4c/5x5_reduce_1[1, 24, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4b/output_1[1, 512, 14, 14], %inception_4c/5x5_reduce_w_0[24, 512, 1, 1], %inception_4c/5x5_reduce_b_0[24])
[v2] %inception_4c/5x5_reduce_1[1, 24, 14, 14] = Relu(%inception_4c/5x5_reduce_1[1, 24, 14, 14])
[v2] %inception_4c/5x5_1[1, 64, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [5, 5], pads: [2, 2, 2, 2], strides: [1, 1]>(%inception_4c/5x5_reduce_1[1, 24, 14, 14], %inception_4c/5x5_w_0[64, 24, 5, 5], %inception_4c/5x5_b_0[64])
[v2] %inception_4c/5x5_1[1, 64, 14, 14] = Relu(%inception_4c/5x5_1[1, 64, 14, 14])
[v2] %inception_4c/pool_1[1, 512, 14, 14] = MaxPool<auto_pad: "NOTSET", kernel_shape: [3, 3], pads: [1, 1, 1, 1], storage_order: 0, strides: [1, 1]>(%inception_4b/output_1[1, 512, 14, 14])
[v2] %inception_4c/pool_proj_1[1, 64, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4c/pool_1[1, 512, 14, 14], %inception_4c/pool_proj_w_0[64, 512, 1, 1], %inception_4c/pool_proj_b_0[64])
[v2] %inception_4c/pool_proj_1[1, 64, 14, 14] = Relu(%inception_4c/pool_proj_1[1, 64, 14, 14])
[v2] %inception_4c/output_1[1, 512, 14, 14] = Concat<axis: 1>(%inception_4c/1x1_1[1, 128, 14, 14], %inception_4c/3x3_1[1, 256, 14, 14], %inception_4c/5x5_1[1, 64, 14, 14], %inception_4c/pool_proj_1[1, 64, 14, 14])
[v2] %inception_4d/1x1_1[1, 112, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4c/output_1[1, 512, 14, 14], %inception_4d/1x1_w_0[112, 512, 1, 1], %inception_4d/1x1_b_0[112])
[v2] %inception_4d/1x1_1[1, 112, 14, 14] = Relu(%inception_4d/1x1_1[1, 112, 14, 14])
[v2] %inception_4d/3x3_reduce_1[1, 144, 14, 14] = Conv<auto_pad: "NOTSET", dilations: [1, 1], group: 1, kernel_shape: [1, 1], pads: [0, 0, 0, 0], strides: [1, 1]>(%inception_4c/output_1[1, 512, 14, 14], %inception_4d/3x3_reduce_w_0[144, 512, 1, 1], %inception_4d/3x3_reduce_b_0[144])
[v2] %inception_4d/3x3_reduce_1[1, 144, 14, 14] = Relu(%inception_4d/3x3_reduce_1

Answer 1 · 2018-10-16T03:55:51.000Z

...
[v3] Initializer runs in 0 ns
[v3] Initializer runs in 0 ns
[v3] Initializer runs in 0 ns
[v3] Initializer runs in 0 ns
[v3] Initializer runs in 0 ns
[v3] InputOperator runs in 1000 ns
[v3] Conv runs in 215357000 ns
[v3] Relu runs in 664000 ns
[v3] MaxPool runs in 36272000 ns
[v3] LRN runs in 17280000 ns
[v3] Conv runs in 22727000 ns
[v3] Relu runs in 126000 ns
[v3] Conv runs in 524730000 ns
[v3] Relu runs in 439000 ns
[v3] LRN runs in 29653000 ns
[v3] MaxPool runs in 25639000 ns
[v3] Conv runs in 14834000 ns
[v3] Relu runs in 33000 ns
[v3] Conv runs in 22210000 ns
[v3] Relu runs in 50000 ns
[v3] Conv runs in 124620000 ns
[v3] Relu runs in 69000 ns
[v3] Conv runs in 3768000 ns
[v3] Relu runs in 8000 ns
[v3] Conv runs in 13229000 ns
[v3] Relu runs in 15000 ns
[v3] MaxPool runs in 26166000 ns
[v3] Conv runs in 7557000 ns
[v3] Relu runs in 16000 ns
Segmentation fault (core dumped)

Answer 2 · 2018-10-16T04:04:15.000Z

I think it's duo to the bug of concat operator.
#100 may fix this. (merged 1 minute ago.)
Could you try again?
<(_ _)>

Answer 3 · 2018-10-16T09:55:55.000Z

Problem solved. Thanks!