Skip to content

Latest commit

 

History

History
71 lines (50 loc) · 3.35 KB

process.md

File metadata and controls

71 lines (50 loc) · 3.35 KB

!!! note "About the process command"

If you have just a few videos to process, or are experimenting, the **process** command is recommended.
This creates a SageMaker Processing Job which uses the [SageMaker ScriptProcessor](https://docs.aws.amazon.com/sagemaker/latest/dg/processing-container-run-scripts.html).
which processes videos one-by-one in sequence. This is the easiest way to get started, but it is not the most efficient or cost-effective way to process videos.
For processing videos in bulk, use the **ecsprocess** command with an elastic cluster.

Process Examples

!!! info inline end The SAGEMAKER_ROLE environment variable is used in the process command. This role is captured in the following order or precedence: 1. Through the SAGEMAKER_ROLE environment variable 2. Through the --config option in the [aws] section or when you run the deepsea-ai setup command

If you are switching AWS accounts, you will need to either set SAGEMAKER_ROLE  or run ``deepsea-ai setup`` again in that new account (recommended).

```
export SAGEMAKER_ROLE="arn:aws:iam::872338704006:role/DeepSeaAI"
```

To run in dry-run mode, which will show you the command that will be run, but not actually run it:

deepsea-ai process -j "DocRickets dive 1423" -i /Volumes/M3/mezzanine/DocRicketts/2022/02/1423/ --dry-run

To run one or more videos in a directory with the job name "DocRickets dive 1423"

deepsea-ai process -j "DocRickets dive 1423" -i /Volumes/M3/mezzanine/DocRicketts/2022/02/1423/  --args "--agnostic-nms --iou-thres=0.5 --conf-thres=0.01 --imgsz=640"

To use a different pretrained model, use the model-s3 option e.g.

deepsea-ai process -j "DocRickets dive 1423" -i /Volumes/M3/mezzanine/DocRicketts/2022/02/1423/ --model-s3 s3://902005-public/models/yolov5x_mbay_benthic_model.tar.gz --args "--agnostic-nms --iou-thres=0.5 --conf-thres=0.01 --imgsz=640"

To process a single video, e.g.

deepsea-ai process -j "DocRickets dive 1423" -i  /Volumes/M3/mezzanine/DocRicketts/2022/02/1423/D1423_20220221T164250Z_h265.mp4  --args "--agnostic-nms --iou-thres=0.5 --conf-thres=0.01 --imgsz=640"

ECS (Elastic Cluster Service) Process Examples

If you have setup an Elastic Cluster Service to process data in batch, you can use it with the ecsprocess command. This is the most cost-effective way to process data in bulk. See ECS Cluster Setup for more information on setting up an ECS cluster with your own pretrained model.

To process videos in a directory with the job name "DocRickets 2022/02" in your cluster called benthic33k :

deepsea-ai ecsprocess -u -c benthic33k -j "DocRickets dive 1423" -i /Volumes/M3/mezzanine/DocRicketts/2022/02/1423/ --args "--agnostic-nms --iou-thres=0.5 --conf-thres=0.01 --imgsz=640"

To process videos in a directory with the job name "DocRicketts 2021/08 with a cluster called mbari315k model", excluding any dives with the name D1371, D1374, or D1375

deepsea-ai ecsprocess -u \
        --job "DocRicketts 2021/08 with benthic model" \
        --cluster benthic33k  \
        --config 902005prod.ini \
        --input /Volumes/M3/mezzanine/DocRicketts/2021/08/ \
        --exclude D1371 --exclude D1374 --exclude D1375 \
        --args "--agnostic-nms --iou-thres=0.5 --conf-thres=0.01 --imgsz=640" \

Updated: 2024-08-14