problems running a retrained tiny_yolov4_license_plates.hef in detection pipeline

General
1

Hi guys

I I retrained a tiny_yolov4_licence_plates with a custom data-set due that one already compiled was for hailo-8 and I have a hailo-8l, I was able to follow all the guide shared in this guide

docker build --build-arg timezone=cat /etc/timezone -t license_plate_detection:v0 .

docker run --name “lpd” -it --gpus all --ipc=host -v /home/tommy/halio_suite/shared_with_docker:/workspace/shared license_plate_detection:v0

./darknet detector train data/obj.data ./cfg/tiny_yolov4_license_plates.cfg tiny_yolov4_license_plates.weights -map -clear

I edited previously files

./cfg/tiny_yolov4_license_plates.cfg:

root@10b08d11ca89:/workspace/darknet# more ./cfg/tiny_yolov4_license_plates.cfg
[net]
# Testing
#batch=1
#subdivisions=1
# Training
batch=64
subdivisions=32
width=416
height=416
channels=3
momentum=0.9
decay=0.0005
angle=0
saturation = 1.5
exposure = 1.5
hue=.1

learning_rate=0.00261
burn_in=1000
max_batches = 2000
policy=steps
steps=3600,3800
scales=.1,.1

[convolutional]
batch_normalize=1
filters=32
size=3
stride=2
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=64
size=3
stride=2
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky

[route]
layers=-1
groups=2
group_id=1

[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=32
size=3
stride=1
pad=1
activation=leaky

[route]
layers = -1,-2

[convolutional]
batch_normalize=1
filters=64
size=1
stride=1
pad=1
activation=leaky

[route]
layers = -6,-1

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[route]
layers=-1
groups=2
group_id=1

[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=64
size=3
stride=1
pad=1
activation=leaky

[route]
layers = -1,-2

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[route]
layers = -6,-1

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[route]
layers=-1
groups=2
group_id=1

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=128
size=3
stride=1
pad=1
activation=leaky

[route]
layers = -1,-2

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[route]
layers = -6,-1

[maxpool]
size=2
stride=2

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

##################################

[convolutional]
batch_normalize=1
filters=256
size=1
stride=1
pad=1
activation=leaky

[convolutional]
batch_normalize=1
filters=512
size=3
stride=1
pad=1
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=18
activation=linear



[yolo]
mask = 3,4,5
anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
classes=1
num=6
jitter=.3
scale_x_y = 1.05
cls_normalizer=1.0
iou_normalizer=0.07
iou_loss=ciou
ignore_thresh = .7
truth_thresh = 1
random=0
resize=1.5
nms_kind=greedynms
beta_nms=0.6

[route]
layers = -4

[convolutional]
batch_normalize=1
filters=128
size=1
stride=1
pad=1
activation=leaky

[upsample]
stride=2

[route]
layers = -1, 23

[convolutional]
batch_normalize=1
filters=256
size=3
stride=1
pad=1
activation=leaky

[convolutional]
size=1
stride=1
pad=1
filters=18
activation=linear

[yolo]
mask = 0,1,2
anchors = 10,14,  23,27,  37,58,  81,82,  135,169,  344,319
classes=1
num=6
jitter=.3
scale_x_y = 1.05
cls_normalizer=1.0
iou_normalizer=0.07
iou_loss=ciou
ignore_thresh = .7
truth_thresh = 1
random=0
resize=1.5
nms_kind=greedynms
beta_nms=0.6

where I changes batch, subdiv, steps mainly

obj/obj.data

root@10b08d11ca89:/workspace/darknet# more data/obj.data
classes = 1
train  = /workspace/shared/licence-darknet/train.txt
valid  = /workspace/shared/licence-darknet/valid.txt
names = data/obj.names
backup = backup/

train.txt and valid.txt it’s the path of the files, something like this

/workspace/shared/licence-darknet/train/00009e5b390986a0_jpg.rf.04d1cbabbde68b6be73ada81c47f3528.jpg
/workspace/shared/licence-darknet/train/00009e5b390986a0_jpg.rf.134d6373e0bde30fc8b9747bc1232667.jpg
/workspace/shared/licence-darknet/train/00009e5b390986a0_jpg.rf.e2f9df03aa702fad603db72632cae9cc.jpg

then I ran this
python …/pytorch-YOLOv4/demo_darknet2onnx.py cfg/tiny_yolov4_license_plates.cfg backup/tiny_yolov4_license_plates_last.weights /workspace/shared/licence-darknet/valid/xemayBigPlate93_jpg.rf.27a32023ca82a9b3685fd6e3833284fe.jpg 1

finally, compilation
hailomz compile --ckpt /local/shared_with_docker/tiny_yolov4_license_plates_1_416_416.onnx --calib-path /local/shared_with_docker/licence-darknet/valid/ --yaml hailo_model_zoo/cfg/networks/tiny_yolov4_license_plates.yaml --hw-arch hailo8lhailo_model_zoo/cfg/networks/tiny_yolov4_license_plates.yaml

Screenshot 2024-12-10 at 12.46.49
Screenshot 2024-12-10 at 12.46.491836×338 129 KB

but when I want to run it in the basic_pipeline detection as the example of barcode im getting some issues, I’m not sure if Im doing it right, but some help it will be good
thanks guys.

NMS score threshold is set, but there is no NMS output in this model.

(Hailo Detection App:7234): GStreamer-Base-CRITICAL **: 12:52:01.565: gst_queue_array_push_tail: assertion ‘array != NULL’ failed

Screenshot 2024-12-10 at 12.52.21
Screenshot 2024-12-10 at 12.52.211920×605 190 KB

2

I didn’t solve it, but I went with another approach, I retrained a yolov8n model with my custom data and it’s working good with the detection pipeline.

3

Hi @tommymcphillips,
Thanks for sharing all the details. I think that you’ve said it somewhere in the original post. The YOLOv4 didn’t have the post-process NMS integrated into the HEF, and the detection example expects that the HEF will include that.