usage: demo.py [-h] [--device {cuda,cpu}] [-m MODEL] [--threshold THRESHOLD] [-i INPUT] [--resize RESIZE]
[--max-image-size MAX_IMAGE_SIZE] [--interval INTERVAL]
demo for object detection
optional arguments:
-h, --help show this help message and exit
--device {cuda,cpu} running with cpu or cuda (default: auto)
-m MODEL, --model MODEL
pre/trained model file (default: torchvision:fasterrcnn_resnet50_fpn)
--threshold THRESHOLD
threshold for accepting detection (default: 0.5)
-i INPUT, --input INPUT
input images, video, folder, youtube url, etc. (default: 0)
--resize RESIZE resize the smaller edge of the image if positive number given, or resize to given size if 2 numbers given
(default: None)
--max-image-size MAX_IMAGE_SIZE
limit the maximum size of longer edge of image, it is useful to avoid OOM (default: None)
--interval INTERVAL detection interval between frames (default: 1)
usage: inference.py [-h] [--device {cuda,cpu,half}] [-m MODEL] [--threshold THRESHOLD] [--max-bbox MAX_BBOX]
[--max-bbox-per-class MAX_BBOX_PER_CLASS] [--nms-thresh NMS_THRESH] [--soft-nms] [--bbox-voting BBOX_VOTING]
[--tta TTA] [--num-dataloader-workers NUM_DATALOADER_WORKERS] [--batch-size BATCH_SIZE]
[--test-window TEST_WINDOW] [--resize RESIZE] [--max-image-size MAX_IMAGE_SIZE] [--auto-contrast AUTO_CONTRAST]
[--clahe CLAHE] [-i INPUT] [-o OUTPUT] [--exclusive EXCLUSIVE] [-j JOBS] [--resume RESUME] [--eval]
[--export EXPORT] [--export-format {Image,Image100,ImageSmallBBox,json}] [--profile]
[--cudnn-benchmark CUDNN_BENCHMARK] [--log-dir LOG_DIR] [--mlflow-tracking-uri MLFLOW_TRACKING_URI]
test script for object detection
optional arguments:
-h, --help show this help message and exit
--device {cuda,cpu,half}
running with cpu, cuda, or float16 (default: auto)
-m MODEL, --model MODEL
pre/trained model file (default: )
--threshold THRESHOLD
overwrite confidence threshold for accepting detection (default: 0)
--max-bbox MAX_BBOX maximum number of bbox output per image if positive (default: 0)
--max-bbox-per-class MAX_BBOX_PER_CLASS
max number of detection for each class (default: 4000)
--nms-thresh NMS_THRESH
threshold for NMS (default: 0.5)
--soft-nms use soft NMS (default: False)
--bbox-voting BBOX_VOTING
threshold for bbox voting, non positive for disabling (default: 0)
--tta TTA Test Time Augementation: orig,hflip,vflip,dflip,autocontrast,equalize (default: )
--num-dataloader-workers NUM_DATALOADER_WORKERS
number of dataloader workers per process (default: None)
--batch-size BATCH_SIZE
batch size (default: 0)
--test-window TEST_WINDOW
size of sliding window for testing (default: None)
--resize RESIZE resize the smaller edge of the image if positive number given, or resize to given size if 2 numbers given
(default: None)
--max-image-size MAX_IMAGE_SIZE
limit the maximum size of longer edge of image, it is useful to avoid OOM (default: None)
--auto-contrast AUTO_CONTRAST
Apply auto contrast to input image (default: None)
--clahe CLAHE Apply Contrast Limited Adaptive Histogram Equalization to input image (default: None)
-i INPUT, --input INPUT
root directory of input images (default: None)
-o OUTPUT, --output OUTPUT
root directory of output (default: None)
--exclusive EXCLUSIVE
txt file contains list of exclusive images (default: None)
-j JOBS, --jobs JOBS number of processes (default: 1)
--resume RESUME path to prediction for resuming (default: None)
--eval evaluation after detection (default: False)
--export EXPORT path of export file (default: None)
--export-format {Image,Image100,ImageSmallBBox,json}
the format of input and output (default: json)
--profile profile the performance (default: False)
--cudnn-benchmark CUDNN_BENCHMARK
enable cudnn benchmark (default: True)
--log-dir LOG_DIR Location to save logs and checkpoints (default: runs/Jun15_02-14-46_dai149)
--mlflow-tracking-uri MLFLOW_TRACKING_URI
URI for MLFlow to which to persist experiment and run data. (default: None)