Hi,
I tried to load a modified GoogleNet (made with CAFFE) with TensorRT2.
I deleted one filter (num_output) in the layer inception_3a/3x3_reduce (first inception module).
I also did the necessary modifications in the caffemodel : All is working fine with CAFFE so I don’t think the problem come from my caffemodel.
Moreover I also tried to delete filters in layers that are not in inception modules and that was working with TensorRT2.
The same Model but without this deletion is also working well with TensorRT2.
Here are my two prototxt parts :
Working Model (with CAFFE and TensorRT) :
layer {
name: "inception_3a/1x1"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "inception_3a/1x1"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>64</b> #HERE
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
Unworking Model (working with CAFFE but not with TensorRT2) :
layer {
name: "inception_3a/1x1"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "inception_3a/1x1"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>63</b> #HERE
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
The error I get is the Following :
genericReformat.cu:1696: genericReformat::NStridesAndOffsets
genericReformat::computeNStridesAndOffsets(const genericReformat::RegionDesc&, const
genericReformat::Array&, const genericReformat::Array&, const genericReformat::RegionDesc&, const
genericReformat::Array&, const genericReformat::Array&): Assertion `(srcStart.i[1] & (1 << srcDesc.lgScalarsPerElement)) == 0' failed.
Aborted (core dumped)
My first thought is that the problem is due to the concatenation Layer but I’m not sure about that
Hi,
We have tested a simple pruned GoogleNet model of 63 convolution filter.
TensorRT can launch it successfully without error.
input: "data"
input_dim: 4
input_dim: 3
input_dim: 224
input_dim: 224
layer {
name: "conv1/7x7_s2"
type: "Convolution"
bottom: "data"
top: "conv1/7x7_s2"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: 63
pad: 3
kernel_size: 7
stride: 2
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
Guess that issue comes from model modification.
Could you re-check your prototxt/caffemodel with Caffe?
Thanks.
Hi,
Thank you for the answer.
I tried your simple pruned GoogleNet model : it is Working Fine
I also tried the following model which is pruned on the three convolution layers it uses but has no inception module and it is also working :
input: "data"
input_shape {
dim: 1
dim: 3
dim: 1000
dim: 2500
}
layer {
name: "conv1/7x7_s2"
type: "Convolution"
bottom: "data"
top: "conv1/7x7_s2"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>62</b> #PRUNED HERE
pad: 3
kernel_size: 7
stride: 2
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv1/relu_7x7"
type: "ReLU"
bottom: "conv1/7x7_s2"
top: "conv1/7x7_s2"
}
layer {
name: "pool1/3x3_s2"
type: "Pooling"
bottom: "conv1/7x7_s2"
top: "pool1/3x3_s2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "pool1/norm1"
type: "LRN"
bottom: "pool1/3x3_s2"
top: "pool1/norm1"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "conv2/3x3_reduce"
type: "Convolution"
bottom: "pool1/norm1"
top: "conv2/3x3_reduce"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>63</b> #PRUNED HERE
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv2/relu_3x3_reduce"
type: "ReLU"
bottom: "conv2/3x3_reduce"
top: "conv2/3x3_reduce"
}
layer {
name: "conv2/3x3"
type: "Convolution"
bottom: "conv2/3x3_reduce"
top: "conv2/3x3"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>191</b> #PRUNED HERE
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv2/relu_3x3"
type: "ReLU"
bottom: "conv2/3x3"
top: "conv2/3x3"
}
layer {
name: "conv2/norm2"
type: "LRN"
bottom: "conv2/3x3"
top: "conv2/norm2"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "pool2/3x3_s2"
type: "Pooling"
bottom: "conv2/norm2"
top: "pool2/3x3_s2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "inception_3a/1x1"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "output"
convolution_param {
num_output: <b>63</b> #PRUNED HERE
kernel_size: 1
}
}
My problem was when I add a pruned convolution layer in a Concatenation layer (For inception module) and the following model is working with CAFFE but not working with tensorRT2
input: "data"
input_shape {
dim: 1
dim: 3
dim: 1000
dim: 2500
}
layer {
name: "conv1/7x7_s2"
type: "Convolution"
bottom: "data"
top: "conv1/7x7_s2"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>62</b> #PRUNED HERE
pad: 3
kernel_size: 7
stride: 2
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv1/relu_7x7"
type: "ReLU"
bottom: "conv1/7x7_s2"
top: "conv1/7x7_s2"
}
layer {
name: "pool1/3x3_s2"
type: "Pooling"
bottom: "conv1/7x7_s2"
top: "pool1/3x3_s2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "pool1/norm1"
type: "LRN"
bottom: "pool1/3x3_s2"
top: "pool1/norm1"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "conv2/3x3_reduce"
type: "Convolution"
bottom: "pool1/norm1"
top: "conv2/3x3_reduce"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>63</b> #PRUNED HERE
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv2/relu_3x3_reduce"
type: "ReLU"
bottom: "conv2/3x3_reduce"
top: "conv2/3x3_reduce"
}
layer {
name: "conv2/3x3"
type: "Convolution"
bottom: "conv2/3x3_reduce"
top: "conv2/3x3"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>191</b> #PRUNED HERE
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "conv2/relu_3x3"
type: "ReLU"
bottom: "conv2/3x3"
top: "conv2/3x3"
}
layer {
name: "conv2/norm2"
type: "LRN"
bottom: "conv2/3x3"
top: "conv2/norm2"
lrn_param {
local_size: 5
alpha: 0.0001
beta: 0.75
}
}
layer {
name: "pool2/3x3_s2"
type: "Pooling"
bottom: "conv2/norm2"
top: "pool2/3x3_s2"
pooling_param {
pool: MAX
kernel_size: 3
stride: 2
}
}
layer {
name: "inception_3a/1x1"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "inception_3a/1x1"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: <b>63</b> #PRUNED HERE
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_1x1"
type: "ReLU"
bottom: "inception_3a/1x1"
top: "inception_3a/1x1"
}
layer {
name: "inception_3a/3x3_reduce"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "inception_3a/3x3_reduce"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: 96 #NOT PRUNED HERE
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.09
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_3x3_reduce"
type: "ReLU"
bottom: "inception_3a/3x3_reduce"
top: "inception_3a/3x3_reduce"
}
layer {
name: "inception_3a/3x3"
type: "Convolution"
bottom: "inception_3a/3x3_reduce"
top: "inception_3a/3x3"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: 128 #NOT PRUNED HERE
pad: 1
kernel_size: 3
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_3x3"
type: "ReLU"
bottom: "inception_3a/3x3"
top: "inception_3a/3x3"
}
layer {
name: "inception_3a/5x5_reduce"
type: "Convolution"
bottom: "pool2/3x3_s2"
top: "inception_3a/5x5_reduce"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: 16 #NOT PRUNED HERE
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.2
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_5x5_reduce"
type: "ReLU"
bottom: "inception_3a/5x5_reduce"
top: "inception_3a/5x5_reduce"
}
layer {
name: "inception_3a/5x5"
type: "Convolution"
bottom: "inception_3a/5x5_reduce"
top: "inception_3a/5x5"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: 32 #NOT PRUNED HERE
pad: 2
kernel_size: 5
weight_filler {
type: "xavier"
std: 0.03
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_5x5"
type: "ReLU"
bottom: "inception_3a/5x5"
top: "inception_3a/5x5"
}
layer {
name: "inception_3a/pool"
type: "Pooling"
bottom: "pool2/3x3_s2"
top: "inception_3a/pool"
pooling_param {
pool: MAX
kernel_size: 3
stride: 1
pad: 1
}
}
layer {
name: "inception_3a/pool_proj"
type: "Convolution"
bottom: "inception_3a/pool"
top: "inception_3a/pool_proj"
param {
lr_mult: 1.0
decay_mult: 1.0
}
param {
lr_mult: 2.0
decay_mult: 0.0
}
convolution_param {
num_output: 32 #NOT PRUNED HERE
kernel_size: 1
weight_filler {
type: "xavier"
std: 0.1
}
bias_filler {
type: "constant"
value: 0.2
}
}
}
layer {
name: "inception_3a/relu_pool_proj"
type: "ReLU"
bottom: "inception_3a/pool_proj"
top: "inception_3a/pool_proj"
}
layer {
name: "inception_3a/output"
type: "Concat"
bottom: "inception_3a/1x1"
bottom: "inception_3a/3x3"
bottom: "inception_3a/5x5"
bottom: "inception_3a/pool_proj"
top: "output"
}
I still get the following error :
genericReformat.cu:1696: genericReformat::NStridesAndOffsets
genericReformat::computeNStridesAndOffsets(const genericReformat::RegionDesc&, const
genericReformat::Array&, const genericReformat::Array&, const genericReformat::RegionDesc&, const
genericReformat::Array&, const genericReformat::Array&): Assertion `(srcStart.i[1] & (1 << srcDesc.lgScalarsPerElement)) == 0' failed.
Aborted (core dumped)
Pruned Inception Module seems to be the problem ?
Regards,
Loïc
Hi,
We have tested the model shared in #3.
TensorRT engine can be launched successfully without the error you meet.
cp -r /usr/src/tensorrt/ .
cd tensorrt/samples/
make
cd ../bin/
./giexec --deploy=<i>/path/to/prototxt</i> --output=output
Looks like this issue comes from the caffemodel you have edited.
Could you recheck if there is anything incorrect when modifying the caffemodel?
Thanks.