Sorry for late reply. Actually I cannot reproduce high memory issue. I ran in Jetson Nano board which is flashed/installed via Jetpack4.5.1.
Below is my step when I run inference with the official release unet model.
It can run inference well against the 720p.jpg file.
Step:
$ git clone
https://github.com/NVIDIA-AI-IOT/deepstream_tlt_apps.git
$ cd deepstream_tlt_apps
$ wget https://nvidia.box.com/shared/static/i1cer4s3ox4v8svbfkuj5js8yqm3yazo.zip -O models.zip
$ unzip models.zip
$ wget https://developer.nvidia.com/cuda102-trt71-jp45 && unzip cuda102-trt71-jp45 && chmod +x cuda10.2_trt7.1_jp4.5/tlt-converter
$ ./cuda10.2_trt7.1_jp4.5/tlt-converter -k tlt_encode -p input_1,1x3x608x960,1x3x608x960,1x3x608x960 -t fp16 models/unet/unet_resnet18.etlt -e models/unet/unet_resnet18.etlt_b1_gpu0_fp16.engine
$ ll -sh models/unet/unet_resnet18.etlt_b1_gpu0_fp16.engine
73M -rw-rw-r-- 1 nvidia nvidia 73M Jul 15 18:25 models/unet/unet_resnet18.etlt_b1_gpu0_fp16.engine$ export CUDA_VER=10.2
$ make
$ ./apps/tlt_segmentation/ds-tlt-segmentation -c configs/unet_tlt/pgie_unet_tlt_config.txt -i /opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.jpg
For 720p.h264, it stops at “NVMEDIA_ENC: bBlitMode is set to TRUE”. But the memory usage is not high.
$ ./apps/tlt_segmentation/ds-tlt-segmentation -c configs/unet_tlt/pgie_unet_tlt_config.txt -i /opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.h264
===== NVMEDIA: NVENC =====
NvMMLiteBlockCreate : Block : BlockType = 4
H264: Profile = 66, Level = 0
NVMEDIA_ENC: bBlitMode is set to TRUE
I also try deepstream_python_apps.
$ cd /opt/nvidia/deepstream/deepstream/sources
$ git clonehttps://github.com/NVIDIA-AI-IOT/deepstream_python_apps.git
$ cd deepstream_python_apps/apps/deepstream-segmentation/$ python3 deepstream_segmentation.py dstest_segmentation_config_semantic.txt /opt/nvidia/deepstream/deepstream/samples/streams/sample_720p.jpg output
The output folder contains the inference result.