Command Line Interface¶
Everything VIAME does can be run from a terminal through the viame command. It suits batch processing, machines without a display, and scripted workflows. The launch scripts in the example and project folders call the same command.
Setup¶
In a desktop installation, load the VIAME environment into the terminal first:
# Linux and Mac
source /path/to/viame/setup_viame.sh
REM Windows
call C:\path\to\viame\setup_viame.bat
When VIAME was installed with pip install viame, the command is available without this step.
Usage¶
viame <command> [arguments]
viame help lists every command, and viame <command> --help shows the options of one.
Commands¶
Running and evaluating models¶
| Command | Purpose |
|---|---|
run |
Process videos or images, or run a single pipeline file |
runner |
Run a pipeline file |
train |
Train detector or tracker models |
monitor |
Monitor a training run and report progress by log or email |
ensemble |
Learn fusion parameters for ensembling multiple detectors |
score |
Score detection and tracking results against groundtruth |
plot |
Plot detection counts per frame, or evaluation results |
segment |
Add SAM2 segmentation polygons to an existing box-level annotation set |
Working with files¶
| Command | Purpose |
|---|---|
convert |
Convert annotation, calibration and registration files between formats |
csv |
Perform filtering and analysis actions on VIAME CSV files |
json |
Perform filtering and analysis actions on DIVE and COCO JSON files |
extract |
Extract frames from video files |
resample |
Resample object tracks from one frame rate to another |
inspect |
Identify a file, check it is intact, and say how VIAME can use it |
metadata |
Dump unified per-image survey metadata for a site folder |
Search¶
| Command | Purpose |
|---|---|
index |
Build and manage the video search index: add, remove, build, list, status, hash |
database |
Initialize, start, stop and index the descriptor database |
search |
Launch the video search (query) interface |
Stereo, registration and 3D¶
| Command | Purpose |
|---|---|
calibrate |
Estimate stereo calibration from calibration target images |
rectify |
Rectify a stereo image pair using calibration parameters |
disparity |
Estimate disparity between a pair of rectified images |
depth |
Estimate depth from a pair of rectified images |
register |
Register survey imagery and detect previously-observed regions |
mosaic |
Stitch a mosaic from images and their homographies |
3d |
Build a 3D model from UAS imagery |
Pipelines and configuration¶
| Command | Purpose |
|---|---|
pipeline |
Generate, inspect, validate and modify pipeline files |
pipe-config |
Configure a pipeline |
pipe-to-dot |
Write the layout of a pipeline as a DOT graph |
pipe-gui |
Run pipelines in a simple interface |
configs |
Extract pipeline and training parameters as JSON |
explore-config |
Explore the configuration of an algorithm |
System¶
| Command | Purpose |
|---|---|
gpu |
Check GPU properties of the system |
add-ons |
List installed add-on model packs and download new ones |
Common Tasks¶
Running a pipeline¶
viame run takes a pipeline and the data to process. The pipeline can be a file, or the name of one in configs/pipelines. The data can be a video, an image, a text file listing images, or a folder.
viame run detector_generic_proposals my_video.mp4
A folder of videos or image folders is processed in one call:
viame run -d my_data_folder -p detector_generic_proposals.pipe
A model file can be given in place of a pipeline. It is wrapped in the default detector pipeline, or the frame classifier pipeline for a classifier:
viame run trained_model.zip my_video.mp4
Changing a setting for one run¶
Any pipeline setting can be overridden with -s, named by its process and setting, without editing the pipeline file:
viame run detector_generic_proposals.pipe -s input:video_filename=my_images.txt
Training a model¶
viame train -i training_data -c train_detector_default.conf
viame train --list shows every trainable algorithm. See detector training for the layout of the training data and the available configurations.
Scoring results¶
viame score computed_detections.csv groundtruth.csv --per-class
See scoring detectors and trackers.
Converting annotations¶
The output format is taken from the file extension:
viame convert annotations.csv annotations.json