I trained the yolov4 tiny model using my own dataset with 6 classes. To use this model with deepstream I followed the steps below:
$ cd /opt/nvidia/deepstream/deepstream-5.0/sources/objectDetector_Yolo
$ wget https://raw.githubusercontent.com/AastaNV/eLinux_data/main/deepstream/yolov4-tiny/yolov4_tiny.patch
$ git apply yolov4_tiny.patch
$ export CUDA_VER=10.2
$ make -C nvdsinfer_custom_impl_Yolo
I opened this cpp file.
nvdsinfer_custom_impl_Yolo/nvdsparsebbox_Yolo.cpp
And then I changed the value of the NUM_CLASSES_YOLO constant to reflect the number of classes in my model.
static const int NUM_CLASSES_YOLO = 6;
I didn’t replace the model parameters with my new model parameters in NvDsInferParseCustomYoloV4Tiny(). Because anchors and mask values are same with in function values.
extern "C" bool NvDsInferParseCustomYoloV4Tiny(
std::vector<NvDsInferLayerInfo> const& outputLayersInfo,
NvDsInferNetworkInfo const& networkInfo,
NvDsInferParseDetectionParams const& detectionParams,
std::vector<NvDsInferParseObjectInfo>& objectList)
{
static const std::vector<float> kANCHORS = {
10, 14, 23, 27, 37, 58, 81, 82, 135, 169, 344, 319};
static const std::vector<std::vector<int>> kMASKS = {
{3, 4, 5},
//{0, 1, 2}}; // as per output result, select {1,2,3}
{1, 2, 3}};
return NvDsInferParseYoloV3 (
outputLayersInfo, networkInfo, detectionParams, objectList,
kANCHORS, kMASKS);
}
I compiled edited model.
CUDA_VER=10.2 make -C nvdsinfer_custom_impl_Yolo
I haven’t had any problems up to this point. When I used the model with the darknet detector demo, the results in the video were good and the classes were pretty accurate. But when I use the model on the same video with deepstream, I get wrong results and it usually only finds the car class.
•Hardware Platform : Jetson TX2
• JetPack Version: jetpack 4.4
•DeepStream Version: 5.0
•TensorRT Version: 7.1.3
•yolov4-tiny.cfg:
[net]
# Testing
#batch=1
#subdivisions=1
# Training
batch=64
subdivisions=4
width=256
height=256
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 = 2000200
policy=steps
steps=1600000,1800000
scales=.1,.1
#weights_reject_freq=1001
#ema_alpha=0.9998
#equidistant_point=1000
#num_sigmas_reject_badlabels=3
#badlabels_rejection_percentage=0.2
[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=33
activation=linear
[yolo]
mask = 3,4,5
anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319
classes=6
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
#new_coords=1
#scale_x_y = 2.0
[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=33
activation=linear
[yolo]
mask = 1,2,3
anchors = 10,14, 23,27, 37,58, 81,82, 135,169, 344,319
classes=6
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=1
resize=1.5
nms_kind=greedynms
beta_nms=0.6
attention=1
gaussian_noise=1
#new_coords=1
#scale_x_y = 2.0
•labels.txt:
car
van
bicycle
bus
motorbike
truck
•config_infer_primary_yoloV4_tiny.txt:
[property]
gpu-id=0
net-scale-factor=0.0039215697906911373
#0=RGB, 1=BGR
model-color-format=0
custom-network-config=yoloTinyResources/yolov4-tiny.cfg
model-file=yolov4-tiny_230000.weights
# model-engine-file=yolov4-tiny_b1_gpu0_fp32.engine
labelfile-path=labels.txt
## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=0
num-detected-classes=6
gie-unique-id=1
network-type=0
is-classifier=0
## 0=Group Rectangles, 1=DBSCAN, 2=NMS, 3= DBSCAN+NMS Hybrid, 4 = None(No clustering)
cluster-mode=2
maintain-aspect-ratio=1
parse-bbox-func-name=NvDsInferParseCustomYoloV3Tiny
custom-lib-path=libnvdsinfer_custom_impl_Yolo.so
engine-create-func-name=NvDsInferYoloCudaEngineGet
#scaling-filter=0
#scaling-compute-hw=0
[class-attrs-all]
nms-iou-threshold=0.3
threshold=0.7