Run inference on the cluster

SainsburyWellcomeCentre / crabs-exploration

A toolkit for detecting and tracking crabs in the field.

BSD 3-Clause "New" or "Revised" License

7 stars 0 forks source link

Run inference on the cluster #186

Closed sfmig closed 6 days ago

sfmig commented 5 months ago

Streamline a bit the process of running inference on new videos using a trained model on the hpc.

Similar to #163 and the work done to run training on the cluster.

Including a quick guide if possible.

samcunliffe commented 5 months ago

Creating environment (and writing to /tmp ) running into permissions errors...?

samcunliffe commented 5 months ago

[@sfmig to review the PR]

Inference:

Producing tracks from a video.
Can pass GT to inference to visualise GT vs tracking.
- if you do this you might as well run the evaluation metrics(?)

What should we log? MLFlow?

We don't always have a MOTA value.
If we don't choose to save frames & don't have GT ... we should default to saving the CSV (but it's not necessarily log-able).

samcunliffe commented 5 months ago

Idea:

crabs.tracking

crabs.tracking.inference
crabs.tracking.evaluation

crabs.detection

❓ crabs.detection.inference
- crabs.detection.evaluation

sfmig commented 6 days ago

Implemented in #251 with the new updates to the CLI and used to generate tracked videos.

It would need to be updated once MLflow is implemented for the detect-and-track command (see #254) - but we can close this issue now as that is now captured in #254