Yolo_v4 TLT-V3 custom bounding box parser is not working

samjith888 · April 3, 2021, 6:59am

Trained a yolo_v4 TLTV3 model . The custom bounding box parser is not working properly after integrating into deepstream

tlt-infer output

pgie_config

output-blob-names=BatchedNMS
parse-bbox-func-name=NvDsInferParseCustomBatchedNMSTLT

But Getting wrong bounding boxes in Deepstream

Deepstream output

[Uploading: IMG_1708.JPG…](Deepstream output)

Morganh · April 3, 2021, 4:03pm

Can you share the config file of deepstream?

samjith888 · April 3, 2021, 6:32pm

tlt-encoded-model=yolov4_resnet18_epoch_011.etlt
tlt-model-key=key
model-engine-file=yolov4_resnet18_epoch_011.etlt_b1_gpu0_fp16.engine
labelfile-path=./labels.txt

infer-dims=3;1280;1440
maintain-aspect-ratio=1
uff-input-order=0
uff-input-blob-name=Input

process-mode=1
network-mode=2
num-detected-classes=1
network-type=0


cluster-mode=3
output-blob-names=BatchedNMS
parse-bbox-func-name=NvDsInferParseCustomBatchedNMSTLT
custom-lib-path=libnvds_infercustomparser_tlt.so

[class-attrs-all]
# nms-iou-threshold=0.5
pre-cluster-threshold=0.1

Morganh · April 3, 2021, 11:04pm

Could you please refer to https://github.com/NVIDIA-AI-IOT/deepstream_tlt_apps/tree/master/configs/yolov4_tlt?

samjith888 · April 5, 2021, 5:17am

Hi Morgan ,
Using the same pgie_yolov4_tlt_config.txt. Even getting the error

Morganh · April 7, 2021, 9:24am

To narrow down, can you download sample yolo_v4 model to check if you can run it successfully?

cd deepstream_tlt_apps/
wget https://nvidia.box.com/shared/static/i1cer4s3ox4v8svbfkuj5js8yqm3yazo.zip -O models.zip
unzip models.zip
rm models.zip