Model engine error, deepstream-test1, python bindings

[leo2105]

Please provide complete information as applicable to your setup.
• Hardware Platform : GTX1080TI
• DeepStream version : 5.0
• TensorRT version : 7.0
• NVIDIA GPU Driver Version : 440.33.01
• Issue type : bugs

I am running deepstream_test-1, I followed all of this README to execute on deepstream docker

Input:
python3 deepstream_test_1.py /opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_1080p_h264.mp4

Output:

Creating Pipeline

Creating Source

Creating H264Parser

Creating Decoder

Creating EGLSink

Playing file /opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_1080p_h264.mp4
Error: Could not parse model engine file path
Failed to parse group property
** ERROR: <gst_nvinfer_parse_config_file:1242>: failed
Adding elements to Pipeline

Linking elements in the Pipeline

Starting pipeline

0:00:00.236785586 2784 0x19b2560 WARN nvinfer gstnvinfer.cpp:766:gst_nvinfer_start: error: Configuration file parsing failed
0:00:00.236801600 2784 0x19b2560 WARN nvinfer gstnvinfer.cpp:766:gst_nvinfer_start: error: Config file path: dstest1_pgie_config.txt
Error: gst-library-error-quark: Configuration file parsing failed (5): gstnvinfer.cpp(766): gst_nvinfer_start (): /GstPipeline:pipeline0/GstNvInfer:primary-inference:
Config file path: dstest1_pgie_config.txt

dstest1_pgie_config.txt:

[property]
gpu-id=0
net-scale-factor=0.0039215697906911373
model-file=/opt/nvidia/deepstream/deepstream-5.0/samples/models/Primary_Detector/resnet10.caffemodel
proto-file=/opt/nvidia/deepstream/deepstream-5.0/samples/models/Primary_Detector/resnet10.prototxt
model-engine-file=/opt/nvidia/deepstream/deepstream-5.0/samples/models/Primary_Detector/resnet10.caffemodel_b1_gpu0_int8.engine
labelfile-path=/opt/nvidia/deepstream/deepstream-5.0/samples/models/Primary_Detector/labels.txt
int8-calib-file=/opt/nvidia/deepstream/deepstream-5.0/samples/models/Primary_Detector/cal_trt.bin
force-implicit-batch-dim=1
batch-size=1
network-mode=1
num-detected-classes=4
interval=0
gie-unique-id=1
output-blob-names=conv2d_bbox;conv2d_cov/Sigmoid
#scaling-filter=0
#scaling-compute-hw=0

[class-attrs-all]
pre-cluster-threshold=0.2
eps=0.2
group-threshold=1

So, I have 2 doubts:

• the code automatically generate mdoel-engine-file? because I dont have it in that path.
• About the last error (/GstNvInfer:primary-inference), I had the same one for deepstream-test4, if you can help, thanks

[leo2105]

Actually, I solved the problem but I had to generate the resnet10.caffemodel_b1_gpu0_int8.engine running the make file in /opt/nvidia/deepstream/deepstream-5.0/sources/apps/sample_apps/deepstream-test1

But suppose I just want to run only in python (/opt/nvidia/deepstream/deepstream-5.0/sources/python/), how to do it? because I had to use C files to creat it or I am wrong?

[amycao]

no matter c version or python version, it will generate engine file in the path specified.

[leo2105]

Thats my question, I run this command

python3 deepstream_test_1.py /opt/nvidia/deepstream/deepstream-5.0/samples/streams/sample_720p.h264

and return this error:

Error: Could not parse model engine file path
Failed to parse group property
** ERROR: <gst_nvinfer_parse_config_file:1242>: failed

Obviously, because It cant find .engine file.
So, the python program doesn’t create this file automatically as I see, idk why, unlike of this program (/opt/nvidia/deepstream/deepstream-5.0/sources/apps/sample_apps/deepstream-test1/deepstream_test1), which do create it.

[mchi]

Hi @leo2105,
It’s not because it’s python. The root cause is you give absolute path to the model, i.e

model-engine-file=/opt/nvidia/deepstream/deepstream-5.0/samples/models/Primary_Detector/resnet10.caffemodel_b1_gpu0_int8.engine

When you give it path started with “/”, it will run into below “if (file_path[0] == ‘/’) {” and fail on realpath() function, and then report Error - “Error: Could not parse model engine file path” because the engine file does not exist.

Maybe this is a bug of DeepStream, we will discuss this internally,

Thanks for chasing this issue!

File: /opt/nvidia/deepstream/deepstream-5.0/sources/gst-plugins/gst-nvinfer/gstnvinfer_property_parser.cpp

static gboolean
get_absolute_file_path (
const gchar * cfg_file_path, const gchar * file_path,
char *abs_path_str)
{
gchar abs_cfg_path[PATH_MAX + 1];
gchar abs_real_file_path[PATH_MAX + 1];
gchar *abs_file_path;
gchar *delim;

/* Absolute path. No need to resolve further. /
if (file_path[0] == ‘/’) {
/ Check if the file exists, return error if not. */
if (!realpath (file_path, abs_real_file_path)) {
return FALSE;
}
g_strlcpy (abs_path_str, abs_real_file_path, _PATH_MAX);
return TRUE;
}
…

[17502156337]

Hi,

I found that it still couldn’t support the absolute path.
Have we had it fixed later on?
And I’d like to know what the realpath function is doing?
Thanks.

[kayccc]

Hi 17502156337,

Please help to open a new topic with more details of your issue. Thanks