Please provide complete information as applicable to your setup.
• Hardware Platform (Jetson / GPU)
GPU
• DeepStream Version
DeepStream 6.0
• NVIDIA GPU Driver Version (valid for GPU only)
Driver Version: 470.103.01
Hi.
I am running the following pipeline for license plate recognition:
uridecodebin -> nvstreammux -> primary nvinfer (car detection) -> nvtracker -> secondary nvinfer (license plate detection) -> nvdspreprocess -> primary nvinfer (OCR on tensor input) -> nvmultistreamtiler -> nvvideoconvert -> nvdsosd -> nveglglessink
I added nvdspreprocess element as I would like to do some pre-processing on license plates in the future (perspective correction), as suggested here. I added some changes to gstnvdspreprocess.cpp to change the ROI in runtime to be the detected object from previous model (license plate) instead of full frame. I can see the ROIs fine on the display.
The thing is, the pipeline works fine (exactly as in the deepstream LPR example) until I add the pre-process element. In other words, I get meaningful output on the display, but when I add the preprocess element and change the OCR model to primary model, I get no output from it (no character labels). Even though the custom parser interface for OCR, NvDsInferParseCustomNVPlate, is being invoked and returning meaningful values (I was able to print license plate characters during runtime but not display them).
-
primary infer element config (car detector)
car_detect_config.txt (809 Bytes) -
secondary infer element config (license plate detector)
lp_detect_config.txt (894 Bytes) -
pre-process element config
[property]
enable=1
gpu-id=0
unique-id=3
# should be same as model gie id (don't know why)
target-unique-ids=4
# 0=NCHW, 1=NHWC, 2=CUSTOM
network-input-order=0
processing-width=96
processing-height=48
scaling-buf-pool-size=6
tensor-buf-pool-size=6
# tensor shape based on network-input-order
network-input-shape= 2;3;48;96
# 0=RGB, 1=BGR, 2=GRAY
network-color-format=0
# 0=FP32, 1=UINT8, 2=INT8, 3=UINT32, 4=INT32, 5=FP16
tensor-data-type=0
tensor-name=image_input
# 0=NVBUF_MEM_DEFAULT 1=NVBUF_MEM_CUDA_PINNED 2=NVBUF_MEM_CUDA_DEVICE 3=NVBUF_MEM_CUDA_UNIFIED
scaling-pool-memory-type=0
# 0=NvBufSurfTransformCompute_Default 1=NvBufSurfTransformCompute_GPU 2=NvBufSurfTransformCompute_VIC
scaling-pool-compute-hw=1
# Scaling Interpolation method
# 0=NvBufSurfTransformInter_Nearest 1=NvBufSurfTransformInter_Bilinear 2=NvBufSurfTransformInter_Algo1
# 3=NvBufSurfTransformInter_Algo2 4=NvBufSurfTransformInter_Algo3 5=NvBufSurfTransformInter_Algo4
# 6=NvBufSurfTransformInter_Default
scaling-filter=0
custom-lib-path=../../gst-nvdspreprocess/nvdspreprocess_lib/libcustom2d_preprocess.so
custom-tensor-preparation-function=CustomTensorPreparation
[user-configs]
pixel-normalization-factor=0.00392156862745098
[group-0]
src-ids=0
custom-input-transformation-function=CustomAsyncTransformation
process-on-roi=0
roi-params-src-0=0;0;200;200
[group-1]
src-ids=1
custom-input-transformation-function=CustomAsyncTransformation
process-on-roi=0
roi-params-src-1=350;200;300;100
- primary nvinfer config (OCR) (I already set the input-tensor-meta boolean to true)
[property]
gie-unique-id=4
#operate-on-gie-id=3
#operate-on-class-ids=0
output-tensor-meta=1
process-mode=1
#force-implicit-batch-dim=1
gpu-id=0
model-engine-file=models/LP/LPR/us_lprnet_baseline18_deployable.etlt_b2_gpu0_fp32.engine
labelfile-path=models/LP/LPR/labels_us.txt
tlt-encoded-model=models/LP/LPR/us_lprnet_baseline18_deployable.etlt
tlt-model-key=nvidia_tlt
#output-blob-names=output_bbox/BiasAdd;output_cov/Sigmoid
batch-size=2
## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=0
num-detected-classes=3
#0=Detection 1=Classifier 2=Segmentation
network-type=1
parse-classifier-func-name=NvDsInferParseCustomNVPlate
custom-lib-path=nvinfer_custom_lpr_parser/libnvdsinfer_custom_impl_lpr.so
net-scale-factor=0.00392156862745098
#0=RGB 1=BGR 2=GRAY
model-color-format=0
[class-attrs-all]
threshold=0.5
What could I possibly be missing here? Thanks.