Failed to parse bboxes in FasterRCNN model, TAO trained detector

[pradyyadav]

Please provide complete information as applicable to your setup.

• Hardware Platform: GPU
• DeepStream Version: 6.2
• TensorRT Version: 8.5.2.2
• NVIDIA GPU Driver Version: 525.85.12
• Issue Type: Question

Hi, I trained a FasterRCNN model using TAO and integrated it with deepstream, the deepstream integration gives the following errors:

nvinfer gstnvinfer.cpp:674:gst_nvinfer_logger:<primary-inference-branch-1> NvDsInferContext[UID 9]: Error in NvDsInferContextImpl::parseBoundingBox() <nvdsinfer_context_impl_output_parsing.cpp:59> [UID = 9]: Could not find output coverage layer for parsing objects

nvinfer gstnvinfer.cpp:674:gst_nvinfer_logger:<primary-inference-branch-1> NvDsInferContext[UID 9]: Error in NvDsInferContextImpl::fillDetectionOutput() <nvdsinfer_context_impl_output_parsing.cpp:735> [UID = 9]: Failed to parse bboxes

While searching for a solution I came across this: Using a Custom Model with DeepStream — DeepStream 6.2 Release documentation.

But I am not sure how to configure the code for my purpose.

Following is the configuration file of the model.

[property]
gpu-id=0
net-scale-factor=1.0
maintain-aspect-ratio=0
tlt-model-key=tlt
offsets=103.93900299072266;116.77899932861328;123.6800003051757
infer-dims=3;384;1248
tlt-encoded-model=/root/export_weapon/frcnn_kitti_resnet18_retrain.etlt
labelfile-path=/root/export_weapon/labels.txt
int8-calib-file=/root/export_weapon/cal.bin
input-dims=3;544;960;0
uff-input-order=0
uff-input-blob-name=input_image
batch-size=1
process-mode=1
model-color-format=1
network-type=0
## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=0
num-detected-classes=3
cluster-mode=1
interval=0
gie-unique-id=9
output-blob-names=NMS
output-tensor-meta=0

[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]
pre-cluster-threshold=1.4
## Set eps=0.7 and minBoxes for cluster-mode=1(DBSCAN)
eps=0.7
minBoxes=1
[class-attrs-2]
pre-cluster-threshold=1.4
## Set eps=0.7 and minBoxes for cluster-mode=1(DBSCAN)
eps=0.7
minBoxes=1

I am confused as this is required to do for my TAO trained model or not ?

[Fiona.Chen]

Please refer to the TAO models DeepStream app sample: NVIDIA-AI-IOT/deepstream_tao_apps: Sample apps to demonstrate how to deploy models trained with TAO on DeepStream (github.com)

[pradyyadav]

The model is integrated, but I am receiving an error saying failed to parse bboxes and Could not find output coverage layer for parsing objects

[Fiona.Chen]

There is already FRCNN sample in NVIDIA-AI-IOT/deepstream_tao_apps: Sample apps to demonstrate how to deploy models trained with TAO on DeepStream (github.com)

[pradyyadav]

Thanks, I went through the FRCNN sample and figured out I just had to generate .so file using the C++ code at deepstream_tao_apps/post_processor at master · NVIDIA-AI-IOT/deepstream_tao_apps · GitHub using the makefile in the folder. I used the command make -C /root/deepstream_tao_apps/post_processor and got .so file generated. Which then I gave the generated file’s location in my model configs.

[Fiona.Chen]

Glad to hear that!