# NOTE: Some values might be overriden via ENV vars (check maskcam/config.py)

[face-processor]
# Detections with score below this threshold will be discarded
detection-threshold=0.1
# Only vote mask/no_mask when detection score is above this
voting-threshold=0.75
# Smaller detections (in pixels) will be discarded
min-face-size=8

# Disable tracker to draw raw detections and set thresholds above
disable-tracker=0

[mqtt]
# These are just placeholders, to enable MQTT define these env variables:
# MQTT_BROKER_IP and MQTT_DEVICE_NAME
mqtt-broker-ip=0
mqtt-device-name=0

mqtt-broker-port=1883
mqtt-device-description=MaskCam @ Jetson Nano

[maskcam]
# Alert conditions
# Minimum people to even calculate no-mask-fraction
alert-min-visible-people=1
# More than this fraction of people without mask will raise an alarm
alert-no-mask-fraction=0.25
# More than this people detected will raise alarm despite no-mask-fraction
alert-max-total-people=10

# Time to send statistics in seconds. Set smaller than fileserver-video-period
statistics-period=15

# Time (in seconds) to restart statistics (and the whole Deepstream inference process)
# Set to 0 to disable / 24hs = 86400 seconds
timeout-inference-restart=86400
inference-log-interval=300

# Other valid inputs:
#  - CSI cameras like RaspiCam:
#    -> argus://0
#  - Any file:
#    -> file:///absolute/path/to/file.mp4
default-input=v4l2:///dev/video0

# Output/streaming video resolution. 1024x576 keeps 4k aspect ratio of 1.777
output-video-width=1024
output-video-height=576

# Run utils/gst_capabilities.sh and find video/x-raw entries
camera-framerate=30

# Only used for argus:// inputs
camera-flip-method=0

# Auto-calculate nvinfer's `interval` based on `camera-framerate` and `inference-max-fps`
# to avoid delaying the pipeline. This will override the fixed `interval` parameter below
# E.g: if framerate=30 and max-fps=14,
#      -> will set interval=2 so that inference runs only 1/3 of incoming frames
inference-interval-auto=1
# Set this value to the actual FPS bottleneck of the model. Only used if inference-interval-auto.
# e.g: run the model on a video file (instead of live camera source) to determine model's FPS on your device
inference-max-fps=14

udp-port-streaming=5400
# 2 ports for overlapping file-save processes
udp-ports-filesave=5401,5402

streaming-start-default=1
streaming-port=8554
streaming-path=/maskcam
streaming-clock-rate=90000
# Supported: MP4, H264, H265
# Recommended H264 for stability on video save
codec=H264

# Sequentially saving videos
fileserver-enabled=1
fileserver-port=8080
fileserver-video-period=30
fileserver-video-duration=35
fileserver-force-save=0
fileserver-ram-dir=/dev/shm
# Use /tmp/* to clean saved videos on system reboot
fileserver-hdd-dir=/tmp/saved_videos

# IP or domain address that this device will show in info messages (logs and web frontend, for streaming and file downloading)
# Recommended: use env variable MASKCAM_DEVICE_ADDRESS to set this
device-address=0

[property]
interval=0
gpu-id=0
# Was: 
net-scale-factor=0.0039215697906911373
#0=RGB, 1=BGR
model-color-format=0

# YOLOv4
# model-engine-file=yolo/facemask_y4tiny_1024_608_fp16.trt
# model-engine-file=yolo/maskcam_y4t_1184_672_fp16.trt
# model-engine-file=yolo/maskcam_y4t_1120_640_fp16.trt
model-engine-file=yolo/maskcam_y4t_1024_608_fp16.trt
labelfile-path=yolo/data/obj.names
custom-lib-path=deepstream_plugin_yolov4/libnvdsinfer_custom_impl_Yolo.so

# Detectnet_v2
# tlt-encoded-model=detectnet_v2/resnet18_detector.etlt
# tlt-model-key=tlt_encode
# labelfile-path=detectnet_v2/labels.txt
# input-dims=3;544;960;0 # where c = number of channels, h = height of the model input, w = width of model input, 0: implies CHW format.
# uff-input-blob-name=input_1
# output-blob-names=output_cov/Sigmoid;output_bbox/BiasAdd

num-detected-classes=4

## 0=FP32, 1=INT8, 2=FP16 mode
network-mode=2
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)
# Default: 2
cluster-mode=2
# Skip inference these frames
maintain-aspect-ratio=0
parse-bbox-func-name=NvDsInferParseCustomYoloV4
engine-create-func-name=NvDsInferYoloCudaEngineGet
scaling-filter=1
scaling-compute-hw=1
#output-blob-names=2012

# Async mode doesn't make sense with our custom python tracker
classifier-async-mode=0


[class-attrs-all]
nms-iou-threshold=0.2

# Default: 0.4
pre-cluster-threshold=0.4