Is there a way to shorten time to create engine from model for development?

• Hardware Platform (Jetson / GPU)
Jetson Xavier NX
• DeepStream Version
DeepStream 6.0
• JetPack Version (valid for Jetson only)
4.6.1-b110
• TensorRT Version
8.2.1.8

I am using pruned PeopleNet. To confirm if the program works correctly, I run it and it takes a couple of minutes to create the engine from the model. Is there any way to shorten the time in order to check the result soon? Thanks in advance.

Do you have model-engine-file set in your pgie config? The engine specified by model-engine-file will be created if doesn’t exist, and then it can be reused. Or you can create engine file from tensorrt and use it directly.

@yingliu thank you for your response. Yes, here is the config

[property]
gpu-id=0
net-scale-factor=0.0039215697906911373
tlt-model-key=tlt_encode
tlt-encoded-model=../../../../samples/models/tao_pretrained_models/peoplenet/resnet34_peoplenet_pruned.etlt
labelfile-path=../../../../samples/configs/tao_pretrained_models/labels_peoplenet.txt
model-engine-file=../../../../samples/models/tao_pretrained_models/peoplenet/resnet34_peoplenet_pruned.etlt_b1_gpu0_fp16.engine
input-dims=3;544;960;0
uff-input-blob-name=input_1
batch-size=1
process-mode=1
model-color-format=0
## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=2
num-detected-classes=3
cluster-mode=1
interval=0
gie-unique-id=1
output-blob-names=output_bbox/BiasAdd;output_cov/Sigmoid

[class-attrs-all]
pre-cluster-threshold=0.4
## Set eps=0.7 and minBoxes for cluster-mode=1(DBSCAN)
eps=0.7
minBoxes=1
[class-attrs-1]
# disable bag detection
pre-cluster-threshold=1.0

in …/…/…/…/samples/models/tao_pretrained_models/peoplenet directory, I don’t have resnet34_peoplenet_pruned.etlt_b1_gpu0_fp16.engine. How do I create and keep?

Is there any *.engine file created in the engine directory?

@yingliu I checked the directory, but there is not…
Here is the list in tao_pretrained_models directory.

xxx@xxx-desktop:/opt/nvidia/deepstream/deepstream-6.0/samples/models/tao_pretrained_models$ ls -lha
total 68K
drwxr-xr-x 17 root root 4.0K  5月 20 06:15 .
drwxr-xr-x 10 root root 4.0K  5月 10 05:42 ..
drwxr-xr-x  2 root root 4.0K  5月 10 05:43 dashcamnet
drwxrwxr-x  2 root root 4.0K 10月 11  2021 dssd
drwxr-xr-x  2 root root 4.0K  5月 10 05:44 facedetectir
drwxrwxr-x  2 root root 4.0K 10月 11  2021 frcnn
drwxr-xr-x  2 root root 4.0K  5月 20 06:17 peoplenet
drwxr-xr-x  3 root root 4.0K  5月 10 05:44 peopleNet
drwxr-xr-x  3 root root 4.0K  5月 10 05:44 peopleSegNet
drwxrwxr-x  2 root root 4.0K 10月 11  2021 retinanet
drwxrwxr-x  2 root root 4.0K 10月 11  2021 ssd
drwxr-xr-x  2 root root 4.0K  5月 10 05:44 trafficcamnet
drwxrwxr-x  2 root root 4.0K  1月 15  2021 unet
drwxr-xr-x  2 root root 4.0K  5月 10 05:43 vehiclemakenet
drwxr-xr-x  2 root root 4.0K  5月 10 05:43 vehicletypenet
drwxrwxr-x  2 root root 4.0K 10月 11  2021 yolov3
drwxrwxr-x  2 root root 4.0K 10月 11  2021 yolov4

Please make sure you run the app with root privilege.

The engine should be generated in opt/nvidia/deepstream/deepstream-6.0/samples/models/tao_pretrained_models/peoplenet/ after the first time running.

Thank you sooo much! As you told, I ran the command with sudo and the engine file was created in the directory, and the program became to start quickly.

This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.