Issues with running inference-workflows process-video

[beeonbike]

Search before asking

• I have searched the Inference issues and found no similar bug report.

Bug

Hi,
I am running inference workflows process-video with a local, docker-based inference server, but unfortunately:

1. with no MP4 output file

in the target directory, I only find 1 evergrowing file, which is not corresponding to the size of the input files.
/nunki/yatina/output/video_output/source_0_output_output_image_preview.mp4
/nunki/yatina/output/video_output/workflow_results_source_0.csv

2. a couple of user warnings
   InferenceExperimentalFeatureWarning: init_with_workflow is experimental: Usage of workflows with InferencePipeline is an experimental feature. Please report any issues here: https://github.com/roboflow/inference/issues
   UserWarning: Field name "schema" in "WorkflowsBlocksSchemaDescription" shadows an attribute in parent "BaseModel"
   UserWarning: Specified provider 'CUDAExecutionProvider' is not in available provider names.Available providers: 'AzureExecutionProvider, CPUExecutionProvider'
   UserWarning: Specified provider 'OpenVINOExecutionProvider' is not in available provider names.Available providers: 'AzureExecutionProvider, CPUExecutionProvider'
   UserWarning: Specified provider 'CoreMLExecutionProvider' is not in available provider names.Available providers: 'AzureExecutionProvider, CPUExecutionProvider'
   Processing video... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━╸ 100% 0:00:04
   Processing: magellanes-cam3_00_20250321053720.mp4

Note: the resulting file is not written!

Environment

Docker version 27.5.1, build 9f9e405
inference version: v0.47.0
inference-cli version: v0.38.0

Minimal Reproducible Example

command for directory

import subprocess
from pathlib import Path

Define the input and output paths

video_dir = Path("/home/yatina/ftpmount/cam3/2025/03/21")
output_dir = Path("/nunki/yatina/output/video_output")
output_dir.mkdir(parents=True, exist_ok=True)

Common inference parameters

workspace_name = "yatinanet"
workflow_id = "aves-time-in-zone"
api_key = "mykey"

Loop through all MP4 files in the directory

for video_file in video_dir.glob("*.mp4"):
print(f"Processing: {video_file.name}")

subprocess.run([
    "inference", "workflows", "process-video",
    "--video_path", str(video_file),
    "--output_dir", str(output_dir),
    "--workspace_name", workspace_name,
    "--workflow_id", workflow_id,
    "--api-key", api_key,
    "--max_fps", "6",
    "--allow_override",
    "--debug_mode"
])

Additional

No response

Are you willing to submit a PR?

• Yes I'd like to help by submitting a PR!

[PawelPeczek-Roboflow]

I am not sure if I understand, but I feel like the command works as it should, yet from what I see, the chances are you iterate over input files keeping the output dir being the same - hence the output files, which have arbitrary names, override each other.

Alternative solutions:
a) create sub-dir for each video file
b) Use the following function directly (not subprocessing to command line entrypoint) - and it's returned value will let you to do whatever you want with results (full paths to output files are provided, you may rename / move etc)

from inference_cli.lib.workflows.video_adapter import process_video_with_workflow

[PawelPeczek-Roboflow]

no further activity - closing the issue